Bit of bad news here I'm afraid - it makes my BASIC interpreter go slower )-:
The test I was hoping to speed up is its sprite handling - it's all done in SDL using the SDL_Blit functions - which are effectively memory move functions. My generic sprite test animates 100 64x64 sprites on an 1280x1024 console at a rate of about 27 fps. With the LD_PRELOAD set for the new memcpy, the framerate drops to 23.
It's no biggie for this, but thought I'd let you know.