[alanbernstein]

I confess I've never used this in a situation where time complexity was at all relevant. For one thing, it is possible that a result only needs to be computed once for each N (and per shape, since this approach works on a wider range of shapes than just spheres).