Ah, there you go. 3DMark03 does not stress the card. If you want an accurate picture of your temps you can either do the 3DM01se nature torture loop or you can run 05 over and over again disabling the menus and repeating of course.
WTF? Please explain how 3DMark03 does not "stress the card", but 3DMark01/05 do? They might use shaders, etc. in different ways (which means that sometimes you'll see artifacts in one but not the others), but they all put an essentially 100% load on the fixed-function rendering of the GPU. Any differences in heat output between different benchmark programs are going to be fairly small.
This is like saying that Prime95 does not "stress your CPU", but Folding@Home does. They may use different parts of the CPU, or put more or less load on the memory subsystem, but the amount of overall heat generated by either is pretty damn close to the max that the CPU can put out. The only way to get closer to the theoretical maximum, realistically, is to have inside knowledge of the design of the CPU/GPU, and to write a program that executes worst-case code in terms of heat output. This usually involves looping over a handful of instructions that don't do anything useful when run like that, and so will never happen in real life.
One other question what does upping the AGP voltage in the bios really do or affect...?
It raises the voltage being supplied to the electronics on your card through the AGP slot. Increasing this is usually a really, really bad idea, as it can damage your card and is unlikely to help increase an overclock. This is NOT the same as modding a card to increase the GPU/RAM voltage (which is also dangerous, but at least might help with stability at higher-than-stock speeds).