My PM also reminded me, depending on the data set size too, if it's bigger than the buffer cache, we might see some improvements over PG as well from that (we've made memory improvements around that which might help potentially).