Looking through my results, I'm not sure if running 4x is more efficient than 1x. With variable run-times I would really need a much bigger sample in order to get some relevent averages.
I did find one shared task with a known setup.
@Egilman at least for this one task, a 480 in Linux isn't quite as fast as it could be. Generally speaking, for most non-DP tasks a 480 should be at least 50% faster than a 7970/280x.
The CPU time is interesting though. I see virtually zero CPU overhead when running 4x tasks in Linux. The biggest spike in CPU usage, that I have tracked, is 3% usage while running 4 XANSONS tasks.