I love experiments like these, that push the limits of what I thought possible with existing technologies. Using Git to version-control the database makes sense, and although the implementation might not be practical/scalable in its current form, it points to an interesting direction.