There's always going to be lots more work to get done, but I think I'm pretty much okay with where OpenGL_accelerate is for now.  There's now a C-level API for format handlers, so that Cython modules can be written that pretty much do a cast and a dereference to get values, rather than doing lots of Python-call overhead.  Don't see much of a speedup in my own code, because it tends to already be doing what its supposed to, passing big arrays of data.  Still, it does help.

There's still lots of testing to do, I've only run on one machine with Python 2.6 and Ubuntu, will need to get Python 2.5, Windows, OS-X etceteras checked out, and then do extensive testing to be sure it's all working.  But that can wait for some spare weekend.