Correction, the 3 way export took only 16 minutes, not 26 minutes.....
I am assuming that this was CS5.5, is that correct?
Render files are typically not used for encoding unless you specify them. What you should do to see if the GPU is doing anything is to turn the Hardware accekeration off to see the difference between GPU hardware assist and just software.
GPU disabled is the same as having no effect supplied. So some how the GPU is being used if I apply an effect (even with no difference being generated, e.g. curves left at their default) in ways I was given to understand it can't, e.g. MEPG2 encoding.
If the GPU doesn't speed up MPEG2 encoding, why the sudden drop in time when in reality that's all that is happening (no resizing etc).
The only reason I was deleting the render files was to make sure there was nothing being grabbed unexpectedly. I know that they are not used, but I wanted to make sure they didn't even exist.
I am going to try an experiment here and will get back later. There are a few strange things about CS5.5 and MPEG encoding.
Well I ran my test on the PPBM5 MPEG2-DVD timeline and tried your three-way color corrector effect and it did not change the time. Now that is not conclusive proof of anything. Our MPEG2-DVD timeline is already loaded with effects both GPU supported and unsupported ones. Sorry I could not find any correlation.
Sounds like it's using Quicktime server to process when there are no effects applied.
with ProRes 422 footage
Eric brings up a good point. Try your test again with camera original media. (Anything not in an MOV container.)