Actually being able to use CUDA and/or offscreen rendering is not possible right now afaik, so getting clarification that it should be possible is a step in the good direction
and I'm looking forward to get details from Aaron and/or Nvidia about this
(ie. how to enable it with current drivers and/or have to wait for a minor bug-fix release about this)
Actually given this 'workaround', being able to at least use the GPU for offscreen rendering (via pbuffers? FBOs?) might even handy enough to start implementing an Optimus-like Mesa 'proxy' driver that accelerates (through Nvidia GPU) select GLX contexts only... which is still better than nothing when it comes to OpenGL development even if then necessarily 'manual/software' framebuffer copy (from offscreen surface to 'onscreen') won't give optimal performance it will probably be still much better than Intel-only in both performance and features.