When I said using this method, I meant writing to backbuffer and copying to texture. Doing that, and using CopyRect, things didn't speed up much. Whereas writing to backbuffer, locking buffers and using CopyPixelFast I got a considerable speed improvement. I haven't tried setting the VRAM flag. I might give that a go, but at the moment I'm considering the program finished and am working on the website now, so tweaks are very low on the agenda.