The temperature measurement of a mainboard can sometimes be very mistaking (even when reading the temperature from the CPU's tempdiode).
Last week, I build a system based on a MSI K7NG2-ILSR mainboard (I have the same board -same revision-, with the same bios, only an other CPU but both with internal tempdiode) and experienced the same strange thing: as soon as I got to stressing the CPU, the temperature of the system got lower.
I can only think of good one reason (that didn't apply to me btw) causing this, and that is a temperature controlled CPU fan (CPU gets hotter, fan accelerates). But I guess that would only keep the temperature down for a little while and it would get up again..
Does this perhaps apply to you?