Check out if there's a BIOS option called something like "CPU Internal Thermal Control". This is the exact name used by Asus. It is found in the BIOS of P4P800 series mobos under "CPU Configuration". You MAY have something similar, or perhaps nothing at all, depending on your mobo & BIOS.
If you have this feature, & it is set to "enabled" or "auto", it will cause a temporary freeze, but not a reboot. This IS NOT the TM1 thermal throttling that has been around for the last several years in P4's, & usually occurs at ~72 - 75*C.
Rather, it is a form of "On Demand" that the mobo manufacturer sets, & is generated by the mobo's BIOS. On Asus P4P800 series S478 mobos it occurs around ~5 - 7*C lower than the P4's internal thermal throttling mentioned above.
Here's how you can check it out if TM1 or On Demand is invoking in WinXP:
Open the tskmgr
Select the 'Performance' tab
Under 'view':
set Update Speed to 'high'
set Cpu History to 'one graph per cpu'
tick 'Show Kernal Times' (this last one is VIP)
Leave the tskmgr open, just as it is.
If you use Prime95 and don't have enough RAM for 2 concurrent torture tests :
Launch the Prime95 window
Pull down 'options'
Select 'torture test'
tick 'custom' on the newly opened window
Enter the desired memory amount in the now available "Memory To Use" box .
Now run two instances of prime 95 (you can do this by running one instance out of two seperate folders w/different names). Or run two instances of any other Distributed Computing client or burn-in ustility that will require 100% cpu utilization continuously.
With both apps launched, the cpu usage will shoot to 100% & the 'Cpu Usage History' graphs will show a continous green line across the very top of the graph. If you open an app, scroll, or move some windows around, you should see some small, brief, blips, of a red graphical line, rising from the bottom of the CPU Usage History graphs. This is your kernal usage, & it should generally be at the floor of the graphs, & very negligible to invisible, when the dual concurrent tests are utilizing your cpu 100%, and you are not doing anything else on the PC, like opening apps, scrolling the mouse, etc.
Now here's where you can detect if the mobo's invoking On Demand. The temps will climb, and then you will see the red kernal line make a very large, wide, spike, most of the way up the graph, while the mouse (and most everything else) *freezes* for several seconds. Then the system becomes 'normal' again (including the graphs), only to repeat the process again within several seconds or a minute, or so. If so, then you've experienced an On Demand thermal event. The system will not crash or reboot, it will simply freeze for a few seconds, maybe even 15 seconds to half a minute, and then resume normal operations, and may then cycle back & forth, with the freezes lengthening, and the normal operation periods shortening, or nearly ceasing.
This sounds very similar to what your friend might be experienceing, but he shuts down the sytem during the thermal event (it is pretty scary the first time you experience it).
If this is happenning, and you can find & disable the feature in your BIOS (or possibly it's tied to the alarm s/w), the testing I mentioned above will probably go a few degrees higher, & then trigger the normal TM1 "thermal throttling". In this case, the green and red lines on the graphs will converge towards the middle (50% usage), and will be all squiggly looking, and going up & down haphazardly. The machine will continue to function, but the proc will process work more slowly, even though speed monitoring utilities will show the proc at its rated speed.
I hope I'm wrong, but I tend to agree w/your friend: It does sound like overheating. Prescotts (especially C0 steppings, and to a slighly lessor extent, D0 steppings) can be a BIT** to cool, and sensors can be way off. I doubt my stock C0 stepping Prescott 3.2 using a stock Intel HSF could handle dual Prime Torture while in a case, w/o tripping my Asus' thermal set-points. But, it does sound like you've got excellent case cooling, so this is somewhat confusing.
I'd first pull the side covers off the box, & see if the prob still occurs at ambient temps. Maybe even set-up a window fan blowing into the side of the case w/the sidecover off. I'd then try a R & R of the stock HSF, using a good TIM, like AS5, & making sure it's well seated. Lastly, I'd consider an aftermarket HSF (XP-90/120, Zalman 7700cu, etc.).
If heat isn't the issue, then the other area to look at might be the PSU. Prescotts should have a min 20A on the 12V rail, IME. But a PSU prob usually results in a reboot, so I don't think that's your prob.
HTH
Strat