However, in many typical graphics-heavy cases, your statements aren't really true. Many of the heavy visual effects are offloaded to the GPU if at all possible (especially in the case of Unreal Engine's OpenGL ES path, as it has to rely on the speed of mobile CPUs otherwise), and data is preprocessed wherever possible so that it doesn't have to be done at runtime. This takes the CPU out of many of the bottlenecks where the difference between asm.js code and regular JITed js would make a large difference. If you're using Emscripten with typed arrays at all, you're going to get a huge reduction in GC pressure as well.
You will be able to scale things like physics (since something like PhysX doesn't run on mobile GPUs anyway), but any gameplay-affecting physical interactions won't be relying on actual CPU-based physics simulation.
In any case, Emscripten is a powerful tool even without asm.js, and I will bet that these demos will still run pretty well in browsers without it. Sufficient incentive (like, a popular WebGL game) would be enough to get framerates as rocksteady as what we're seeing in Firefox with support.
The big news here is that Mozilla has convinced Epic to bet on the web (where they were previously throwing all their weight behind their Flash-based engine), and they've demonstrated with a major, non-trivial piece of software how easy the porting process can be. Moreover, because it's Mozilla, we know that they'll make sure that cross-browser support is baked in, so that if or when other browsers are fast enough to play these games, they'll be able to.
Serious kudos to Mozilla for this win.