Not being an expert on the topic, I once needed to come up with a simple algorithm to cluster locations on a map, where every point needed to be part of a cluster, and clusters were bound in several ways (not just in size (i.e. point count), but also in some other measures which depended on varying factors).
By first finding the "difficult" points, and then growing clusters from there, I got the best results. "difficult" points would be points that are hard to reach, e.g. because they would lie alone in the outskirts of the total area, or because they would help hit another cluster boundary condition more than other points. This helped to grow neatly aligning clusters, leaving very little loners and corresponding handwork to place them.
This may help you if your current algorithm would normally find these difficult points last.
