Hi Ivan and thank you for your question.
There are certainly ways we can look to tune the scenario you describe above. The first thing I’d be looking to do is to mock this up in as simple a way as possible and profile it using insights. That will give you a good idea of where the load within the physics tick is coming from. (What I mean here is that it is no good optimizing the broadphase, only to find you are transferring the load to the narrowphase section, which would still need running to check if the collision still exists). Once we have that we can take a stab at methods to reduce the time spend on the tick.
In a more generic answer - there were some earlier physics algorithms which worked well with temporal coherence - Sweep and Prune exploited this pretty well. However as worlds have become more dynamic the update cost tended to dominate the saving of temporal coherency.
All the best
Geoff