Leaving DRAM at 1.65 and lowering VccIO to 1.15 would be 0.5 difference, so I would try that first. Intel have warnings and stuff for using more than 1.5V ram though, So I would maybe also try to run the ram at 1.5 and increase the timings. I imagine setting it to 9-9-9-25-2 may work better with lower volts... worth a shot, but ugly if there is no profile for it.
VccIO should be within 0.3 and 0.5 Volts of DRAM Volts. Greater than 0.5 diff and I read the is too much potential difference between the memory controller and the DRAM (not sure whether that is a correct explanation though). The spec for VccIO is also 1.05V +-3%, but most people ignore this part.
Also VccSA should really be under 0.971 volts but these MB use VccIO supply for VccSA so even more reason to get low volt ram... My VccIO is 1.01875 or something so well under 0.3 diff from 1.25V. I am not sure what going under 0.3 does, maybe it causes some difficulty for the processor to distinguish the memory controller in the die or something, but it seems to work fine. I was going to try 0.971 when I get really bored, might try tonight.
This is just going from specs though and the specs are just specs, with no explanation.
Ivy Bridge might be better with low volts so if u planning on that I would ditch that ram ASAP and get 1.25V DRAM.
If you are happy to keep it try booting with MemOK pressed to see what it does, write down all the values then set manually.
You might find VccIO to 1.15 makes Sandy happy, and still keeps 1.65V in spec of 1.65-1.15=0.5V difference. It is very borderline though. Maybe also 1.175 VccIO or thereabouts would work if you find the RAM is being overvolted by the board slightly. (or just undervolt the ram yourself to counter the board)
2133 is not really different from 1600 as these chips are real world limited by instruction cache, not MHz, GHz
EDIT: 0.971 worked for VccIO/SA for 1.25V ram!... sorry added it here because as usual I rambled on above and someone might wonder what the result was.