The GPUs are not specifically being leveraged by MATLAB.
With the Radeon cards, MATLAB just submits everything to the graphics drivers. The exact path to the graphics drivers and the rendering speed can depend upon the figure renderer that you have in effect; you might want to change that.
If you were doing 3D work then probably your graphics would be handed over to your system OpenGL drivers which would be responsible for optimizing GPU usage.
The situation is not actually much different for NVidia cards, but the Parallel Toolbox at the very least knows how to speak to CUDA version 1.3 and later, so hypothetically MATLAB could call upon CUDA for increasing rendering speeds. I have never heard of it actually doing that, though. Most of the NVidia cards do not reach CUDA 1.3 standards. For example, none of the integrated NVidia graphics cars for Apple systems can do CUDA 1.3 (but there are two cards available... provided you have a G5 Tower.)
Anyhow... MATLAB itself can't speak directly to the GPU in your Radeon, and even if you had a suitable souped-up NVidia card, current MATLAB versions would not speak directly to the GPUs for rendering assistance. Some day, maybe...