The reason JPG compression shouldn't have an effect on *in memory use* is that once the texture is loaded, it is loaded into memory *decompressed*. That way we don't have to do decompression on the image every time the texture is accessed during rendering (which would create needless rendering overhead). Instead we just have direct access to the per-pixel color, with no decode step. That's my understanding anyway. In other words, no matter how compressed an image is (even if you use max JPG compression), the *in memory* size should be basically the same.