Why is there any point in trying with faster memory if you are only at 80% perf on DDR3? You'll just go to 40% perf on the next iteration?

You should only upgrade when the bottleneck becomes memory, right now the bottleneck is still on compute until you are at 100% max perf