One thing I noticed when working on a prototype to a virtual display driver (same problems, if
you can't write directly on the graphic card frame buffer), was that if you have a big latency
but have enough bandwidth, it may be cheaper to do the deferred write to all the pages,
instead of controlling what different pages changed.
In the case of a virtual driver, if you have a modern graphic card on the host side, it will
not be noticeable in speed if you do 640x480, 800x600 or a 1024x768 blit, but it will be
noticeable if you do a sequence of smaller blits to changed parts of the screen (making it