Consider a set of documents with two equal sized classes that are split into two clusters.
If there is one cluster that contains few points and one cluster that contains
almost all the points, then the entropy of the resulting solution is very
nearly the entropy the larger cluster