I was (am?) one of the two people who ran a ton of benchmarks (game benches, realtime) to test the effect of RAM speed on framerate and frame time. I have a 3930K + RIVE so while I can't test 2600+ speeds, I am able to disable cores and run tests simulating a quad core or dual core, and I tested everything from single to quad channel, with and without HT, and with a single 7970LTG, single 680LTG, single 670FTW, 2x 670FTW, and 3x 670FTW. I tested resolutions ranging from800x600 tto 2560x1600, but the vast majority were in 1080p.
1333 @ 6-7-6, 7-7-7, 8-8-8, and 9-9-9
1600 @ 6-7-7, 7-8-7, 8-8-8, and 9-9-9
1866 @ 7-8-8, 8-8-8, 8-9-9, 9-9-9, 9-10-9, and 10-10-10
2133 @ 8-10-8, 9-10-9, 9-11-9, 9-11-10, 10-11-10, 10-10-10, and 11-11-11
2200 @ same as 2133
2400 @ 8-11-10, 9-10-9, 9-10-10, 9-11-10, 10-11-10, 10-10-10, and 11-11-11
The short conclusion is this:
Higher mmemory speeds results in higher minimum frame rates almost universally.
Max FPS is affected only slightly except when CPU bottlenecked (SC2 with 400 units on screen)
Gains of 5.4 to 27.1 FPS min were seen
Gains of 0 to 13 FPS max were seen
Gains of 4.1 to 15.8 FPS avg were seen
That's going from 1600 8-9-8-24 to 2133 9-11-10-28
The difference increased by about 10-22 percent when you increase to 2400 9-11-11
There was no situation in which higher memory speed decreased framerate
People who use multiple GPU setups can expect to see a difference of 2.9 to 41.8fps min (Nvidia: 9.1-34.9, AMD 2.9-41.8)
The game play is very noticeably smoother, with the amount of stuttering decreased by a power of 2 (Nvidia users: zero perceptible stuttering, AMD users: greatly diminished stuttering in CFX)
I've been testing even further using a delidded 3770K @ 5.1Ghz on a Gigabyte GA-Z77X-UP7 with memory speeds that include all prior tested as well as 2600 9-12-12, 10-12-12, 11-13-12, 2666 @ same as 2600, 2800 10-12-12, 10-13-11, 11-12-11, 12-13-13, and 3000 11-13-13, 10-14-12, 11-14-12, 12-13-12,and 13-13-13. So far the results are identical percentage wise.
So, in my oopinion, the best RAM for the money is either 2133 CL9 or 2400 CL9. Both give identical results (9-11-10-28 vs 9-11-11-31) and both are available right around $100 for 16GB and 8GB kits respectively.
If you choose 1600, even with timings that are unreachable for most memory like 6-7-7-17, you are between 2.8-and-17.3 percent slower than the off-the-shelf 2133/2400 kits.
I didn't expect quite so much difference, but I also tested everything at least 5 times (most tests were done 7 times and the two outliers removed), and I continue to test and will continue for quite some time, hopefully finishing and being able to get the results and writeup published.
My advice is to buy 2133 with the tightest timings possible, as it's the best deal in terms of performance per dollar you can get.