Yep Tumor, the 128MB 8500 is faster than the 64MB 8500. It appears ATI moves some more of the driver functions to the hardware of the video card and uses less AGP memory for textures, which would improve the speed a bit.
Considering running in 1280x1024x32 bit color requires 5MB of video memory for any given frame, I can see why the speed is faster. They can fit the multiple screen images (back buffers) in video memory and still have plenty of scratch-pad ram left to work the pipelines.
Heck, it only takes 8MB of video ram to hold a 1600x1280x32 bit frame. They can stuff multiple frames in video ram on this card easily and still have scratch pad ram available.
This prevents stalling of a game engine, which leads to potentially higher frame rates and smoother action.
While the RAM requirements for a frame seem low, a lot of video ram is used for the operating software module of the card. They used to run the operating module out of a ROM, but this is just too slow, so they load the card's RAM with the operating module. This is also why you do not want to run with video BIOS caching enabled on the motherboard.