If too many membership requests time out in the MembershipQueue, it can cause the event loop to lag.

[Half-Shot]

Basically if too many requests take too long to handle (i.e. a thundering herd of startup traffic), then the membership queue can become saturated and eventually start to fail requests. If these requests are not suitably spread out, the TTL timers can all fire at once and cause the bridge to slow while it tries to process all of them.

The answer is probably to do less things in the membership queue, spread out the operations, and see if we can stop using timeouts for TTLs (as oppose to just expiring them silently).