We found some issues with thermal throttling on the SB81P when we reviewed it in July, and despite the passage of time, the trouble doesn't appear to be completely corrected. You will find my exposition of the conundrum right here. Mike saw the same sort of throttling problems I did, although he says a newer BIOS helped a little.
As he worked with the SB81P and observed lots of throttling, Dr. Schuette also noticed an additional hitch caused by the activation of the CPU thermal throttling mechanism in the Pentium 4. As the CPU throttles back and becomes less available to process requests, system memory gets abused:
The problem is, however, that even though the addresses are generated, the data cannot be output on the bus because there is no buffering capacity left on the chipset. The consequence is a series of retry (commonly described as a loop) with the side effect that the memory pages are kept open for an extended period of time for read accesses. Keep in mind that during an interleaved read, the power consumption of memory chips is about 10x that of the same device in active standby (no activity). The thermal dissipation follows the same rules.So when the XPC SB81P busts loose with a fit of CPU thermal throttling, system memory heats up. This behavior isn't specific to this Shuttle box, I believe, but it does occur often on the SB81P.
More troubling still, Dr. Schuette raises the issue of the delay between an increase in fan speed and the onset of additional cooling introduced by the heat pipe cooling system in the Shuttle box. This cooling setup may not be well suited for Prescott processors that can change temperatures very quickly. He speculates that the system may need to be reengineered in order to handle a Prescott CPU in combination with "smart" fan speed controls without some throttling.