Assessing the speedup due to parallelization