Yeah, the video is exclusively a test of physics performance, rather than the general aspects of the tool. Admittedly the performance of a physics engine in Javascript is a weak point. FWIW, we've been experimenting with using the native physics engine integrated in to CocoonJS, which could speed it up a great deal on there. They used that on the video so it would help their test, but browsers would still be slow.
I've actually experimented a little with using an asm.js build of Box2D. On Firefox it's way faster - literally 2-4x faster. However the asm.js code was slower on Internet Explorer and a few mobile devices I tried, presumably because they don't actually support asm.js so the overhead increases even more. If lots of mobile devices supported asm.js, I'm sure we could even make physics in the browser really fast.