It looks like there are three pillars of "eyefinity" that need to be combined for the monitor wall seen in the videos.
enough outputs on each card
A method to spread the drawing over multiple GPUs
My understanding is that 3. (and any kind of multi-GPU-on-single-X-server support) is a huge TODO item that can't be picked up until kernel memory management is done and may not be started until after the G3D transition.
bridgman, when you said it's probably easy to support in OS drivers, did you refer to points 1. and 2., or did you include point 3.?