The rotations of the cameras are a bit too slow though, but mainly because of our setup, so this can be a lot better. Our custom protocol couldn't really handle dropped packets, so if we filled the sender's or receiver's buffers, the servos would start to spasm. To counter that we just didn't send head position as often as we could.
Of course, by fixing that you still get the round-trip time of turning your head -> moving the cameras -> getting updated image back. But we think it's feasible.
I've always thought fixed, super-wide-angle lenses and software would be the ideal way to go - the groundstation does all the work.
And the video is made for showoff, the project in itself is about the applications of a system like this in a specific industry.