Javascript is not the problem (though I think we would all be better off with more variety like NaCl would have provided).

I'm actually starting to wonder if an approach like docker/lxc, but with web features ported to native land may not end up making more sense in the long term.

The idea would be to take all the features of the web that aren't document focused and make them available in userland in a linux container. For example, the linux container would have to ask for permission from the host operating system to get GPS access or Contacts API access. It would also support the ability to use URIs to manage the state of any chromeless "window" object.

It may be easier to take the web to sandboxed containers than bring all the operating systems concepts to the browsers. Basically this would be a LXC User Agent.