The main aprs.he.fi server was feeling a bit sluggish - it was starting to run out of memory, and it did some swapping. This happened simply because there are so many people looking at the maps at the same time, and fetching real-time updates from the server at a rather high constant rate. So today I separated the web server to another server, and left the database running on the old server computer. This seems to help the situation quite a lot, and did not require a lot of work.
There was a very short outage caused by the change - by accident the new frontend server did not serve your requests for a couple minutes, since I had accidentally forgotten an access list in it's configuration. It only worked for me! Sorry about that.
Maybe in the future I'll try to put together a couple of dedicated frontend servers and start load-sharing web requests between them. Adding a slave database server would make the HA setup almost completely fail-safe. The software I've written seems to support scaling to a cluster with multiple servers just fine. I had hoped it would, since it was designed to, but I didn't try it until now.