So I have an inventory system which features physics actors for items that are on the ground. When picked up, they are set not to simulate. When dropped, they are turned back on and set awake.
To rule some stuff out - it happens even if CCD is off on the item physics actor, and both the item physics actor and the level geometry are simple collision boxes or capsules.
What happens is - if all the items are sleeping, the performance is normal (100-120 FPS depending on what’s on screen). If an item is dropped, the performance instantly goes down to about 50FPS. Waking multiple physics actors doesn’t seem to decrease this further linearly, so even if you drop 5 items, it stays with that 47-50FPS range.
Literally the moment the actor (or all awake actors) goes back to sleep, the framerate jumps back to normal 100-120FPS.
I haven’t tried breaking the test out into a separate project yet or anything, but it should be easy to reproduce by making a static mesh with collision, turning it into a physics actor, and waking it and observing a perf hit.