QueryPerformanceCounter bug or chipset bug?

So we've been wrestling with a problem that seems to be with QueryPerformanceCounter. Consider the following test code.

#include <iostream>
#include <Windows.h>

int main(const char **_args, int _argn)
{
	LARGE_INTEGER frequency, lastTime, currentTime;
	QueryPerformanceFrequency(&frequency);
	QueryPerformanceCounter(&lastTime);

	std::cout << "Freq: " << frequency.QuadPart << " : counter " << lastTime.QuadPart << std::endl;
	while(1)
	{
		QueryPerformanceCounter(&currentTime);

		LARGE_INTEGER difference;
		difference.QuadPart = currentTime.QuadPart - lastTime.QuadPart;

		double dDeltaTime = (double)difference.QuadPart / (double)frequency.QuadPart;

		if((currentTime.QuadPart < lastTime.QuadPart) || (difference.QuadPart < 0))
		{
			std::cout << "Bad Timer Values: Last Time " << lastTime.QuadPart << 
				" current Time " << currentTime.QuadPart << std::endl <<
				" difference " << currentTime.QuadPart - lastTime.QuadPart << std::endl;
		}
		lastTime = currentTime;
	}


	return 0;
}

Running this on various single cpu machines around the office outputs no errors, however we have a collection of duel core AMD machines that continuously spam out bad values(negative difference). Anyone know anything about this type of thing? We've noticed a post online where apparently this problem can be 'fixed' by setting the processor affinity or something to a specific processor, but I'm concerned that doing this could adversly effect the performance of the application. I just wanted to get some more input. Could anyone out there with a duel core cpu or multi-processor machine run the above code and tell me if you get any output with a negative counter difference. I'm wondering if this is a bug with duel core/multi-proc machines in general, or if its a possible chipset bug. Thanks J