VivienM: I think the fundamental idea of running Prime95 and some such for hours on end is to test whether or not the CPU, HT bus, memory, graphics card, etcs' silicons have any defects at the frequency they're running at. If these, what are considered the most intensive tasks one can stress any part of the system, pass for hours on end then most likely the physical properties of the various parts are conforming to the "standards of stability" set forth by their manufacturers (and AMD systems DO pass these tests). This means that any programs that run on these hardware are guaranteed to get the correct results, given that the program itself is correctly written. In the end, if the hardware that the software is running on is correct (not considering cosmic events, such as cosmic radiation, natural disasters, major power fluctuations, etc...), then no blame can be attributed to the hardware itself. That being said, if Windows still crash/implodes, then that's Window's problem, be it drivers or general uptime.
Then given both conditions that the drivers are correct in addition to the hardware itself being correct (both most likely the case being that Prime95 et al. passed), then the fault can only be attributed to Windows' own overall stability. In this case, then you would experience the same, if not longer, uptime on the AMD system than the Intel system (since you have the optimal hardware/software driver configuration, which is also probably the case for your old P4).