I'm using ATI's Compressinator, without much luck -- I seem to get major artifacting. In the above image, I took the original (left, 3.0MB), then saved as JPEG (center, 0.2MB). I then took the original and saved as DDS DXT5 (right, 1.0MB).

The on-disk savings wasn't as much as I'd hoped (though the in-memory savings is nice), and I'm getting artifacts up the wazoo. Am I missing something?

As a sidenote, we were playing around with something that allowed us to use JPEGs and alpha transparencies (essentially one JPEG for RGB, and a second greyscale JPEG for the alpha). It worked pretty well, but of course, only saved disk space; not video memory.