Why use different kinds of static point clouds? If you have a one that's dense enough and a projector with enough resolution, can't you display any 3D model to as much accuracy as your resolution allows? Or are the different point clouds tailored to certain kinds of models?

Really cool. I really want to play that 3D maze-like game.

The link says this was created in 2006/7. Have there been any advances since then in terms of resolution, or otherwise?

Not directly, but there was some work on animating speech in virtual avatars[1] (constructed from a single image of a face) which uses the cubes to make animated 3d avatars.

What I found especially interesting is that even though the point-etching in the cubes is static (and of a generic head), faces are mostly similar enough (and the animations required for speech small enough) that looking at the animations actually looks very realistic in person.

[1] http://www.cs.columbia.edu/CAVE/projects/avatar/