Fixes done in 3.4.20 to introduce 2 threads for event serialisation seem to have moved a performance bottleneck further downstream on busy instances. Looks like the rule executors can now cause excessive DB load, and cause high db load.
Lets try to optimise 2 things here:
Calls for queue stats (currently iterates over a lot of items)
Zipping issue payloads to reduce network & disk I/O