This paper here: http://www.cs.unc.edu/~cyfu/pubs/dac11.pdf provides insight into how using multi-cores can increase parallelism through minimizing synchronization as a bottleneck.