This is really interesting, though we would have to check our assumptions before applying it. For the N-body problem that I think this is referring to, we can only be confident about our boxes if planets follow the general rule of being either well clustered together and not varying too much in mass (which I think makes sense in practice?) as a planet that's 10 grids away from some center but 4 orders of magnitude larger than average would be much more relevant in computations than average-sized planets one grid away.

I think one way around this may be to first calculate each planet's effect on other planets using this method with the number of grid cells being dependent on planet size. That could also help save some computation when dealing with planets that are too small to effect others.