I'm still not sure I believe that USB 2.0 is faster than an internal IDE directly on the system bus. I just don't see that happening. Too many factors against it, from latency due to distance, to number of conductors, to most basic design concepts. Sure, it would be faster than an external hard drive, but internal vanilla IDE? I'd have to see some very credible proof before believing that.
EDIT: FWIW, Wikipedia puts a PATA (aka IDE) drive's practical transfer speed, under average conditions, at 50-60MB/second. It also puts a full speed USB (all prior to 2.0 and many 2.0) device, not connected to a hub, at 12MB/second. A hi-speed USB (many 2.0 but certainly not all) has a theoretical max of 60MB/second but a usual performance max of 30MB/second, again only if not attached via a hub. USB devices attached via a hub all share the throughput max of the port to which the hub is attached, so if you have a 4 port hub each with a device, they would be sharing a usual 12MB/second throughput, and even then only ever achieving max individual performance if the USB hub in question has individual transaction translator for each hub port (not common in cheap hubs).
So, assuming he used a good hi-speed (make sure to check your specs, 2.0 is not hi-speed by spec) key, connected directly to its own motherboard port, he will get speeds of a third to half of internal IDE. Given that USB keys are known to slow down as their write cycles are used up, it won't be long before he's looking at a more expensive device with less capacity, much slower speeds, and a very finite lifespan if he uses a USB key as his primary storage device.
A 10g PATA/IDE drive is almost free and suffers none of those limitations.