An interesting variation on this is the VList, which works the same way except the inner arrays have variable size. This gives you an asymptotic rather than constant speedup for many operations.

http://en.wikipedia.org/wiki/VList