So with the launch of Z77 comes this big jump in memory frequency over previous platforms (well previous Intel platforms at least) and the bandwidth gains that come with it.
So far it looks like there are three main schools of thought,
1) Insanely tight timings (7-10 perhaps 6-10/9 or maybe an Andre/Shamino special 5-9) with memory frequencies between 2400-2700
2) Reasonable timings (9-11/12 10-11/12) with frequencies 2600-2800
3) Loose timings and high frequencies 2800+
Let me just weigh in on my person thoughts real quick.
(1) Insanely tight timings and high frequency will give you the best result, but it does in most cases require LN2 on the memory and brings with it high risks. If you are going after the 3DMark01 or SuperPI 32m world record, yes this is the best option, but not for every day benching. This would be done with PSC or BBSE.
(2) The ideal option for most users. A good mix of bandwidth, reasonable timings and frequency will be optimal for most benchmarks. The risk is also very low with this option, as you can run this on air.
(3) So far in my testing I havn't found a time when this is the best option. In most cases to run memory this high you have to loosen off the secondary and tertiary (third) timings and you take a bandwidth hit.
When you bench Ivy Bridge, it is more than likely your chip will bench at full pot or close to full pot. Due to these cold temperatures and the fact your cold boot bug will require you to continually warm up/cool down means condensation/water will probably gravitate towards your memory slots, just like on AMD. Andre/Shamino need some serious props for running their memory on subzero for all their benching, that seriously isn't easy and adds another level of risk/troubleshooting into the benching. That aside, most benchers will want to run memory on air as it will be much easier to manage especially when ice/water is already impacting the memory slots, which is why for 99% of benching I choose options (2).
GSKILL have put out their TridentX series of kits, all of which look very strong. I have tested the F3-266611D and Dino is testing the 2600 and they both have strong performance. Here is some results that I have been able to achieve with probably the strongest memory clocker around the ASUS M5G.
GSKILL F3-266611D rated at 2666 11-13-13-35 2N
ASUS ROG Maximus 5 Gene Z77
Intel 3570K E1
Combination of air cooler and single stage (not really any noticeable memory clocking difference)
These sticks are designed specially for that (2) I was talking about, reasonable timings and good frequency. Please keep in mind on this platform TRCD doesn't really make much of a bandwidth difference, tightening CAS can be good for 32m, but timings as you might be familiar with them on Sandy Bridge/Gulftown act in totally different ways.
Stock frequency, tightened up with 10-12-12-31 1T
1350 MHz, 10-13-13-31 1T
1400 MHz 11-13-13-31 1T, 1.7v
1423 11-13-13-31 1T, 1.74v
1443 11-13-13-31 1T, 1.74v, slightly looser subtimings
As you can see from these results, the bandwidth is a little up and down. Some of that can be to do with the CPU frequency, but also some is the tightening/loosening of sub-timings to achieve specific frequencies. While this kit is designed specially to offer option 2 with good timings/good frequency, it also has the super high frequency there if you want it. For me benching day in day out I would be looking to sit it around 1333 10-12-12 or 1350 10-13-13 as this will give me plenty of bandwidth for 3D and also give strong 2D performance.