If we could render with multiple threads, we could use 4 threads to set up all OGL commands for your sprite renderer and then call them in order from the main thread a bit quicker since all the state changes could be optimized for the display list and maybe even processed ahead of time.
@princec : IHMO There should be a opengl context (one per process), which “owns” some number of logical devices (one per GPU…or might show only one in case of things like SLI/Crossfire)…each device own’s it’s textures, queries, programs, etc. etc. Although it would probably be reasonable to allow telling the context that the backing of things like textures and programs are the same. As far I as know you can’t really use two nivida cards (other than Quadro…WGL_NV_gpu_affinity which is a PITA) in opengl individually.
You’d be surprised how far this tactic can get you. I would say if anyone in any position of authority where you work has opinions like these, you are screwed. As Riven pointed out, you are typically screwed. In this case not giving a shit becomes even more important. You get paid the same, the only thing that changes is you don’t have any stress at work.