Some more ideas, both are similar but different way of conceptualising
Pressure Release Valve
The network operates smoothly under ‘normal’ variability but changes behaviour under stress conditions. I think this sudden ‘valve’ style change in network behaviour is not great but it does lead to consideration of what counts as dangerous stress. It also might encourage less sudden changes in client behaviour if clients know their changes might make their experience worse.
Queue
The network processes requests during busy times based on some queue mechanism, like bitcoin does with the transaction fee. In bitcoin the transaction fee is used to put a queue of transactions in some order or priority, and during times of stress when the queue is very long the network can still function predictably.
In SAFE maybe the queue is ordered by the type of event (vault join vs GET request vs vault relocate vs PUT request vs split event vs hop message relay etc) but there may be other factors that contribute to when and why events are processed. What could the network do when the queue gets very long?