One very important aspect in user interfaces is visual responsiveness.  When the user provides input there needs to be some sort of visual (or sometimes audial) response that clearly conveys to the user that it acknowledges the input and that it is doing (or already has done) what has been requested of it.  Although modern systems are very fast, visual response is still limited to the sum of the time that the window server can communicate with the client program, the time it takes for the program to create the visual response and then for the window server to composite the result.  In Vision, the visual response is not subject to as much of the communication latency because the drawing occurs as soon as the event hits the window server and gets published.