The problem is that you need to translate TGSI back into something the host machine can understand and execute. Systems that can interpret TGSI directly are virtually non-existent right now on Linux (and completely non-existent on Windows and Mac OS X hosts).
On Windows and OSX, Gallium3D as a whole is completely nonexistent. The only way Gallium3D is going to be useful for virtualization, is if hosts expose Gallium3D, including TGSI, directly. That's an assumption I took, because without it, the Gallium3D option isn't even available.
I really, really wish I had the expertise to write a KMS and Gallium3D driver myself. All I can do is offer to test code. Any other ideas? I'm open to writing them a letter on paper, if it will help.
Do you know OpenGL? Then there's nothing stopping you from joining the Gallium3D project. The Gallium "pipe driver" API is quite similar to OpenGL, so if you know the former, you understand the latter too. And that's the whole story. And I disagree that working on Mesa/Gallium and drivers is hard - it isn't. Reading someone else's code is the only way to get in, as is the case with any other project. I used to work (and am working now as that's what I am paid for) on projects which are much harder to work with and so much bigger pain in the ass that Gallium looks like an easy deal, thanks to its great architects.