It would seem that the algorithm frequently deals with nodes with a length of 1, which is inefficient. Nodes should be merged whenever possible.