Quote - Seems to me that all video cards display OpenGL accelerated textures in powers of two. That is, 2,4,8,16,32,64,128,256,1024,2048,4096, up to 8192.
If image files are anything in between those resolutions, the video card has to internally resize the image to display it. Assuming I'm right, that is.

The optimization is for memory usage and not speed.
A texture will occupy the same amount of memory as the next higher power of two size.
There is no resizing of images and thus no difference in speed. However more textures can fit inside the video card's memory if they are in power of two.