Raytrix has already been doing this: http://raytrix.de, however stereo-cams, time of flight, and structured light approaches are usually more effective. You can do everything you can do with a light-field array camera with 3+ lenses and some clever algorithms.
Correction: you can do everything Lytro is doing with lightfields with 3+ lenses and some clever algorithms. Lightfields are incredibly useful rich 4d datasets with massive potential (would you like a side of automatic shader generation with your 3d scan?) but Lytro is totally blowing it by focusing (no pun intended) on this trivial little aspect of it.
