I’m not sure if it’s fair to dismiss ARC as a bad fit for server-side work in general, but the typical argument is that atomic retain/release operations become quite expensive when you share memory across threads. It’s easy to imagine, for example, how a global database connection pool accessed on every request across dozens of threads would have non-zero ARC overhead.

In practice, compiler optimizations reduce the amount of retain/release pairs emitted, but I have no idea how the resulting performance compares to GC languages.