Doesn't this just push the problem up the stack one layer? Now the W3C needs to define a specification for a virtual machine that can implement all of the current web specifications performantly. If your VM is missing an API to perform some new form of user input (like multitouch) or a rendering method for a new standard (like webgl) you'll need to head back to a standards body to add more hooks, leaving us in the same position.

I think the idea has merit, but the current landscape of software development highly favours the statically compiled C++ browser engines, especially on mobile devices.