Every major site overloaded and lots of hardware upgrades.

Sites are running multiple Docker images of Lemmy-server on the same underlying hardware.

This code change show how much of an issue having the MTA in-process is impacting things: https://github.com/LemmyNet/lemmy/pull/3466

The client-facing API of Lemmy_server also would benefit from caching layer, that server operators could control - even if only for emergency.