Hi,
In the attached file, there is a simple PhysX interaction. It worked fine in 029 or 030 version. In 033 the nakelace pass through the hook.
I could raise PhysX settings but then sim is slower...
You're not using PhysX. You're using the bind solver. Nothing about the bind solver changed in the last few versions. The difference you're seeing could be due to the fact that you have strict determinism turned off, so successive sims vary. I loaded your scene with v0.16029 and the beads still fell through the hook.
The bind solver has a natural tendency to pull particles past narrow surfaces due to the fact that it does not yet consider edge-face intersections for bindings themselves...instead it only considers vertex-face intersections of particles against colliders. You can try several things to alleviate this. For example, with a small timestep like you have, you can reduce the bind solver step size (higher the step size, the easier it is for large strings of bindings to pull small groups of bindings through narrow colliders). Also experiment with reducing the collision compensation mass multiplier (try .25) and increase friction on your hook collider.
You can also try an actual PhysX-based approach, which will be more accurate, but also slower.
