Maybe I don't understand how MW works... you explicitly elect not to receive CPU tasks, but you receive them anyway, and different ones than mine with the inverted selection.

No Sek, I currently run only GPU at MW (my current pref's I posted). The stat's I quoted was when I was running both CPU and GPU there and wanted to test the difference.

A GTX260-216 has 216 shaders, so selecting to use 384 shaders or 216 is just the same. Selecting 128 would be a mistake; if that selection button still does something.

CC1.3 cards (200 series) are double precision, but do not perform as well as ATI cards on MW. CC1.2 and CC1.1 cards such as the GT240 and 9800 GT are not capable of running MW tasks as they cannot compute in double precision.
Also, the NVidia 400 series cards are 8% efficient while the top ATI cards are 40% efficient at MW.
At GPUGrid some of the tasks are 98% efficient on my GTX470s (I have two in an i7-920 and it draws about 490W at the wall, shaders OC'd to 715MHz).

So ATI is better suited to MW and NVidia are better suited to the much more complex single precision calculations at GPUGrid.

I would turn my i7 Fermi rig off before I ran MW on it; Fermi's had double precision all but stripped out.

I wondered about that MW setting and why it would affect the user if set higher, when GPU crunching really should only be done when you're not using the computer. Does anyone crunch on GPU during use? It's one of the deeper concerns that WCG has... the impact on the volunteer.

Snooze is something not offered on Linux other than by running a boinccmd instruction and I'd imagine that if you have multiple cards in a system, you could tell one to use for the display and the others for crunching... then you could even run the 3D BOINC screensaver ;P