Thanks all for the tips.
This is a Windows XP machine. Resetting the affinities in task manager worked to open up the other core. The GPU clients were both set to run on CPU1 so I set them both to run on both CPUs. Now I have both cores runnning at 100% utilization but, curiously, not much of an improvement in GPU throughput.
This was puzzling - I would not have thought that the two clients would completely swamp the Opteron. I paused one of the clients and the CPU utilization dropped to exactly 50%, spread across the two cores. I resumed the first core, CPU utilization went back to 100%, and then I paused the other core. CPU dropped to exactly 50% again, spread across the two cores.
I don't know how to interpret these results. Is there something in the client code that restricts it to no more than 50% loading, regardless of it's need?
Right now I'm testing the stability of a 25% overclock on the processor. Then I'll run the clients again to see if the utilization percentages change. After that, maybe I'll underclock the GPUs.
This problem was caused by Windows, which was created by Microsoft Corporation.