I'm pretty sure it's a clock issue (a computer's clock is more or less the same as in a quartz wristwatch, where you see it losing or gaining time).

I did several small computer music generation projects many years ago when you basically had to toggle the speaker on/off at the right frequency for each note. You had to determine how long to keep the speaker on and off for each note. So, let's say it's 10 cycles on, 10 cycles off. This works great - assuming the cycles take exactly the same time every time. If the hardware clock "wobbles", going a bit faster or a bit slower sometimes, then the cycles will execute correspondingly faster/slower. Since the cycles are now longer/shorter, the pitch will change. If you run such a program on a faster computer, the pitch will shoot up the program is assuming the clock speed is supposed to be the same.

So, a wobbly clock would explain the subtle pitch shifts I've noticed. If the clock is slightly slower than it's supposed to be all the time, then you get what OncTech describes.

As to what causes the clock discrepancy, it could be the software enTourage has been working on for sleep and low power modes, which might involve the clock being slowed (like Intel's SpeedStep), or it could be the system hardware drivers need a tweak, either by enTourage, or by Marvell if Marvell provided them for their chip.