I certainly understand that going from 0 to 1M polys has to have extreme impact on fps - but only a momentary one. But once I have all the balls in the view, move camera slightly so that they all stay in the view while still having some 70 fps - I would expect the camera movement to be smooth because there are no sudden peaks or drops in the displayed polycount.

But anyway - public tests like this (and hopefully some more complex) are exactly what should have been done by Conitec a long time ago. By doing that there would be no "my fps is suddenly lower" posts appearing with every major update