It's not that much of a claim, if they implicit parallelization works well. Also some functional languages are getting very nice optimizations like deforrestions / fusions that make them really fast.