Except for embarrassingly parallel problems, the trade-off of generating more parallelism is usually needing finer-grained communication. Canonical examples in the literature are matrix multiplication, triangulation, and refinement.