No, you can't for example control RGBW one-wire LED strings with a Raspi. On the BBB thanks to the PRU you can do so with 0% CPU load.
You can run programs with the vector processors in the vidcore with QPULib. https://github.com/mn416/QPULib Kind of apples and oranges, considering the PRU is more focussed on determinate timing.
No, but the GPU actually contains a non-ARM general-purpose processor. I think we're in the very early days of knowing how to program for it.