i don't have any hard links/evidence but it really seems to me that CPU usage should directly depend on which has the best hardware interface. Thus, theoretically, you should be able to achieve about the same CPU usage with any of the three. In practical use, my USB 2.0 external HD is noticibly faster than my 4200 (? whatever the slow laptop standard is) rpm laptop hard drive.
I think it's all in the hardware. Check the [H]'s review of the nforce2 components. As i recall the cpu usage was ~4-10% for usb use, don't remember the rest.
USB 1.1 'wasted' a lot of clock cycles just looking to see if anything was actually plugged in. I know USB 2.0 fixed most of that, allegedly.
The external USB 2.0 drives i've used 'seemed' slower than my internal equivalents, but i have no numbers to back that up with.