I measured one query that does a sequential scan over millions of rows. In PG 9.4 this only uses one core. Current versions are adding a feature to parallelize this. If I had run several queries in parallel, PG would have used all cores.