That's cool that they are trying, but what's the opportunity cost of not doing it? How much effort would it be to simply optimize asm.js now and then undo the optimization later when the core JIT gets smarter?

Seems to me THAT is the way to go, but this is not my area of expertise.