We should stop enumerating subsets before we allocate too much memory. Note that this only applies if we have a large number of nodes. In particular, pre-allocating memory for all subsets according quickly gets expensive.