So, r196 has had some changes, related to this post, I suspect:
Delta-time measurements are now based on the browser's rendering timestamps instead of time measured from JS code. This means motion should be smoother since it is based on when the browser is scheduling frames instead of when the JS code happens to run.
For one dt is suddenly way more accurate (no clipped to 2 digits):
But at the same time - look at that framerate - same capture conditions, there's almost no CPU load when it runs without capture and yet unlike the previous example it never goes over 40 fps. And those DT numbers - are we in sub-tick territory now?