+1.
Mainstream graphics cards outclassed the needs of 2D applications long ago. Most of the things that Photoshop will be slow on are the result of CPU- and RAM-intensive tasks, not display issues. Photoshop will occasionally use the 3D aspect of your video card to do certain transformations, but many photographers and designers will never even use those specific filters in the first place.
When you load up a huge image in Photoshop, it's not going onto your VRAM - only your "screen" does. When you apply a lens blur filter, it's your main CPU that is doing all the complex specialized (and in this case likely proprietary) math, not a generic video card GPU. Even when you zoom in and out of your image, it's not a case of the graphics card scaling a window - that's what happens when you use Exposé, and is why text gets fuzzy in such instances - but the software doing its own interpolation.
It would probably be possible to write some functions as graphics card operations, but that would be pretty stupid as it would be substantially less compatible, require more developmental resources, and likely not increase performance all that much. 3D "acceleration" works because it offloads some very predictable yet intensive math (as well as a lot of low-resolution textures and some 3D models) onto a specialized card, *while* your main program (which is itself resource-intensive) can run on the main system. In Photoshop's case, the program itself is the graphics - there's little point in writing it to run on a graphics card, whose 2D functions are normally limited to scaling, rotating, compositing, color rendering, etc. - when it will run better on the main system.
Does that make sense?
EDIT: I thought of one instance in which it might make sense to write Photoshop functions for the GPU, which would be to take advantage of its highly parallel nature. Of course, these days people are buying multi-core CPUs, which completely negates that. So again, no real point.