Good afternoon.

I notice in my completed tasks that the run time is around 5x the CPU time in duration.

Looking at the top contributors, it appears there is a way to get your run time to nearly equal the CPU time, on the long runs at least.

I've tried changing the app_config to dedicate 1, 2, 3 or 5 CPU units to the GPU task, but I'm getting unchanged results. Does anyone understand how they are accomplishing this, or is it a trade secret?

There should be no trade secrets in non-profit science. The "secret" is something called SWAN_SYNC. If you are running windows you can see how to integrate it here:
http://www.gpugrid.net/forum_thread.php?id=4589#47419

If you are running Linux you can view this:
http://www.gpugrid.net/forum_thread.php?id=4813&nowrap=true#50807


This one parameter allows for up to 20% better performance for this application.