The first synthetic benchmark we used is SiSoft Sandra. This offers a very comprehensive look at the way the memory interacts with the processor. In theory, faster memory gives better cache response, bandwidth and memory bandwidth.
As you can see the G.Skill Trident X 2400MHz kit scaled reasonably well, although if you consider their is also a 2133MHz memory divider that slots in between 1866MHz and 2400MHz then you can see scaling is not perfect. Scaling starts to tail off after 1866MHz and the G.Skill 2400MHz kit shows signs of this in the synthetic benchmarks.
In the SuperPi test, the system calculates 32 million digits of Pi. This is a CPU/memory intensive process so in theory using faster RAM should offer good scaling. Although, we only saw scaling between 1600MHz and 1866MHz. The difference between 1866MHz and 2400MHz was so small it could easily be put down to margin of error.