Hosting a dozen WordPress sites on one server is a wager that none of them will ever fail badly enough to take the rest down with it. The infrastructure decisions that matter are ways of hedging that wager, and most setups make them in the wrong order. They start with caching, because speed is visible and easy to show a client. Separation comes late, or never. The eight decisions below run the other way, from the one that decides how much a single mistake can destroy down to the one that mostly decides how fast the homepage loads.
Making Correct Decisions
The below ranking assumes the kind of server agencies and freelancers actually run, somewhere between five and fifty client sites on hardware they manage themselves. A single hobby blog would order things differently, and so would a platform with its own security team.
1. Every Site Gets Its Own Account
Separation belongs at the top, and on the available evidence it isn’t especially close. On a typical hand-built server every site runs as the same system user, often the web server’s own account. So, a hole in one site’s contact form plugin can hand an intruder the files of every other site on the machine. The odds of such a hole existing somewhere across a dozen installs are not small. Patchstack logged 11,334 new vulnerabilities across the WordPress ecosystem in 2025, 91% of them in plugins. Its data also puts the typical gap between disclosure and the first attacks on the most heavily targeted flaws at about five hours. The company sells protection against exactly this problem, which is worth keeping in mind, but its plugin figures match what anyone who has cleaned up a compromised server already suspects.
The fix is dull work. Each site gets its own system user and its own pool of PHP processes running as that user, with file permissions that keep one pool out of another’s folders. A compromised plugin then compromises one site, which is bad enough. The documentation for cPanel describes PHP-FPM pools as a way to run each website under its own user, and by default it names each pool’s user after the hosting account. That default explains the danger in a common shortcut. A site added as a folder under someone else’s account inherits that account’s user and everything it can touch. The shortcut saves a few minutes on the day the site is created.
2. Where the Hardware Argument Bends
The second decision concerns the machine underneath, and it complicates the case just made. Everything above argues for more walls. A crowded virtual server does offer one, between your collection and everyone else’s, but nobody on your side controls it. Behind that wall you share processors and disks with strangers whose traffic spikes turn into your slow afternoons. Twelve well-behaved sites can still be starved by a neighbor you’ll never meet. Moving them onto one physical machine removes that outer wall and concentrates the risk, which sounds worse, and in one narrow sense it is. What it buys is predictability, since the only neighbors left are sites somebody chose to put there.
Anyone pricing that route could start with something like the spec sheet for Atlantic.Net’s dedicated hosting. Its entry configuration pairs six processor cores with 64GB of memory on mirrored solid state drives, and the machine is standalone hardware rather than a slice of a cloud platform. Root access comes with it, along with a free hand over the operating system and the control panel, and for this purpose that arguably matters more than the core count. Per site users and separate PHP pools only exist on a server where somebody holds the authority to create them.
On sizing, add up each site’s busiest month rather than its average one, because consolidation plans tend to run out of memory well before they run out of processor time. The same cPanel documentation mentioned above recommends enabling per site PHP pools only on servers with at least 2GB of memory, or at least 30MB per domain. Treat that as a floor for the process manager. A single PHP worker serving a WooCommerce store with a page builder on top can use several times that 30MB on its own, and the database and the object cache will want their own share of whatever headroom is left.
3. Multisite Solves Someone Else’s Problem
Third on the list is a question WordPress raises on its own, whether to run everything as one Multisite network or as separate installations. Multisite is tempting on a shared server because it promises one update for the lot. Every site in a network shares the same core files, and although each one can activate plugins and themes, none of them can install either. For a university running forty department blogs under one IT team, that arrangement is the whole point. For an agency hosting a bakery next to a law firm, it largely undoes the first decision on this list, because one shared codebase becomes one shared point of failure. Sooner or later the bakery’s owner will ask for a plugin the law firm’s compliance officer would never allow. Inside a network, that plugin’s code then sits in the shared files under both sites, whether the law firm activates it or not.
When the sites share an owner and an update policy, a network usually earns its keep. When they have different owners or different legal obligations, separate installs tend to cost less over the life of the server, even though they mean more updates to run. Leaving a network later is possible but rarely pleasant, since each site’s tables and uploads have to be pulled out of the shared install and its users untangled from everyone else’s. That makes the first choice worth more thought than the one update convenience suggests.
4. Database Credentials That Stop at One Site
Fourth is the database, and the usual mistake here is laziness passed off as convenience. One MySQL server for every site is usually fine. One database user with rights over every database is a different matter, yet plenty of servers end up that way because someone pasted the same credentials into each site’s wp-config.php file during a busy week and never went back. Each site deserves its own database and its own user, permitted to touch that database and nothing else. A custom table prefix is sometimes offered as a substitute. It hides tables from careless scripts and does nothing against an account that already holds rights over them.
MySQL’s own security guidelines put the principle in a single line, advising against granting more privileges than necessary, and they point to a built-in command that lists which accounts can reach what. On a server hosting a dozen sites, running that check takes minutes. Any account allowed into every database deserves a second look, and so does a password that nobody has changed since the developer who chose it left the agency. If an attacker reads one configuration file, the damage should stop at one set of tables.
5. The Oldest PHP Version Sets the Pace
Fifth, and underrated, is whether every site has to run the same PHP version. On a server with one global version, the oldest and least maintained site sets the pace for everyone else. PHP branches get four years of support in total, two with active fixes and two with security patches only.
PHP version usage data from WordPress.org puts version 8 or later on about 79.34% websites, which leaves more than 20% on version 7, a branch the project stopped patching years ago. WordPress core usually copes with newer PHP without complaint. The trouble tends to come from a page builder or a payment plugin whose last update predates the current branch.

5.1. Upgrading One Site at a Time
Separate versions per site, which the per site PHP pools from the first decision make fairly easy, let the healthy sites move ahead while the stubborn one gets its own deadline. Hosting panels offer the same idea folder by folder, and a walkthrough on changing the PHP version of a WordPress site shows how little ceremony it involves. The difficult part is telling a client that their site is the one holding the whole server back.
Two habits make that conversation easier.
- Keep a plain list of which PHP version each site runs and which plugin, if any, is keeping it there, so the conversation with the client starts from a named plugin rather than a vague complaint.
- Then test the next version on a staging copy before touching production. A page builder that throws warnings on staging is an hour’s work to fix, and the same failure at checkout on a Saturday can cost a client.
6. Backups Belong Somewhere Else
Backups rank sixth, which some readers will find absurd. They’re either the most important item here or the least, depending on whether something has already gone wrong, and this list is ranked for the days before that happens. On a shared server, the location of the copies decides whether they’re worth anything. A backup sitting on the same disk, owned by the same user an attacker has just taken over, is a backup the attacker can delete or tamper with. Attackers appear to know this.
- Plugin schedulers, like the setup covered in a guide to automatic WordPress backups, make a reasonable first layer.
- The second layer belongs on a different machine that pulls copies from this one, so the server never holds credentials that could wipe them.
- Keep enough days of history to reach back past an infection nobody noticed at first.
- Restore one site onto a scratch server every few months as well, since nobody knows whether a backup works until someone has restored it.
6.1. Restoring Onto a Clean Machine
Restores carry a trap of their own. Restoring onto a machine that hasn’t been cleaned can simply hand the attacker their work back. The safer sequence is slower. Stop the affected site’s processes or rebuild the server, restore from the off-server copy, and then change every password the compromised account could have read, database credentials included. Per site accounts pay off again here, because the cleanup can be limited to one account’s processes rather than the whole machine’s.
7. Scheduled Tasks Nobody Scheduled
Plugin driven backups usually run on WordPress’s own scheduler, which leads into the seventh decision, one almost nobody makes on purpose. WordPress runs its scheduled tasks when a visitor loads a page, a design that works on one site and gets unpredictable across twelve. A quiet site can miss its backup for days because nobody dropped by. A busy one can launch several heavy jobs in the middle of a traffic spike, all fighting every other site for the same processors.
The Plugin Handbook’s page on hooking WP-Cron into the system task scheduler describes the way out. Each site’s built-in trigger gets disabled in its configuration file, and the server’s own scheduler calls the site’s cron script on a fixed interval instead. On a shared server, two refinements are worth adding.
- Stagger the calls a few minutes apart rather than firing a dozen sites at the top of the hour.
- Run each one as that site’s own user so a scheduled task can’t reach beyond its own folder.
The job takes an afternoon and tends to be missing from setup checklists.
8. Why Caching Ranks Eighth
Caching comes last, deliberately, and this is where the ranking argues with itself. Page caching and a shared object cache will usually do more for load times than anything above, and on a busy server they may be the only reason the machine stays upright on a Monday morning. They also bring sharing back in through a side door. A single object cache serving every site needs a distinct key prefix for each one, or two sites can end up reading each other’s stored data. Even with prefixes in place, a purge triggered on one site can empty the cache for all of them.
The Redis Object Cache plugin’s installation notes suggest changing both the prefix and the database number for each site to avoid cache data collisions, and the second half of that advice is the easy one to skip. Separate database numbers keep one site’s purge inside its own space, so a plugin clearing its cache after an update doesn’t cold-start every other site on the server at once. Speed work done before separation work has a way of gluing the sites back together.
An Afternoon With the Weakest Site
None of this is exotic, and that’s the uncomfortable part. A lot of servers carrying a handful of WordPress sites were set up in an evening and then grew one site at a time, with nobody revisiting the design because nothing had broken yet. You probably already know which site on yours is the weak one. It belongs to the client who never pays for updates, or it’s the old project nobody wants to touch.
That site is where the next free afternoon should go. The account question comes first, since it decides how far everything else can spread. Find out which user its PHP processes run as and whether that user can read any other site’s folder. If it can, most of the other checks can wait until that’s fixed. After that, look at what its database account is allowed to reach. Then restore its newest off-server backup onto a disposable server. If no such copy exists, that’s the finding. A server built in an evening will often fail at least one of those tests, and the fix for the weakest site usually turns out to be the fix for the rest.
Somebody will find a way into that site eventually, and the vulnerability counts leave little room to hope otherwise. Whether the other eleven learn about it from a log file or from their own customers was mostly settled on the evening the server was built. An afternoon can reopen that decision, though nothing on the server will ever ask for one.





