I can provide screenshots and files when I am back home at the weekend.
Most likely it is not a bug. We are most certainly just doing something wrong in the process. Would be great to figure out what.

Thanks for the input so far.

Edit:
Ive been testing this issue for some weeks meanwhile with different models and different methods, remodeling the surs in different ways for these tests.
And yes, I would say red parts are where the collision lag started in most of my observations. I believe I even have a video showing how it is triggered.