I don't think you could reasonably expect to utilize it to "natively" render stuff like NES, or GB, or any kind of graphics like that. The main problem is that the data PowerVR expects is totally different from what something like the SNES or Genesis produced. Some amount of translation is going to have to happen, and usually that involves the CPU rendering things into a texture format the GPU can understand when we're talking about these kinds of gaming consoles.
Now, if you're just trying to do hardware accelerated drawing (software render a frame, use OpenGL or D3D or whatever to draw it on your computer's screen), this might help if it could somehow detect duplicate groups of pixels and cache them for later reference. But even then, I can't imagine you'd get another more than a minor or negligible boost in speed. I mean, it already doesn't take that much for OpenGL to blit something to a screen with GBE+ (and other emulators too) when it's all just a single textured quad, so you might be hard-pressed to find improvements that are noticeable.
Now, if you're just trying to do hardware accelerated drawing (software render a frame, use OpenGL or D3D or whatever to draw it on your computer's screen), this might help if it could somehow detect duplicate groups of pixels and cache them for later reference. But even then, I can't imagine you'd get another more than a minor or negligible boost in speed. I mean, it already doesn't take that much for OpenGL to blit something to a screen with GBE+ (and other emulators too) when it's all just a single textured quad, so you might be hard-pressed to find improvements that are noticeable.