I had a chance to mess around with voltages and IBT AVX settings last night. Some interesting results...
Settings: First...I upped my volts to 1.40v @3.9GHz with LLC2, from 1.3685v (despite 1.3685v passing IBT AVX preset of Very High). This is where I did my testing - 1.40v, Auto PB, LLC2, 100x39, RAM 3200MHz CL14 1T.
Run 1: The first thing I did was test IBT AVX on Maximum preset. The result took forever so I only did one pass. I got a result value of around 3.5E-2 with a Gflops of 177. Huzzah! Much more consistent with your results.
Run 2: So my thought was, okay, more RAM means higher speeds and bigger result values. So, I did a custom test using 8000Mb of RAM and got an unusually low Gflop of like 142 and a result of like 2.97E-2 or something.
Run 3: My third test was my normal Very High preset which uses 4096Mb of RAM. I was right back to my normal values of ~150 Gflop and 2.99E-2 result.
Speculation of results: I can't say that more RAM means higher values because a custom 8000Mb test yielded pretty crappy results. Maybe because I'm using a non-x64 memory quantity that requires weird behaviors from RAM chips? However, I can conclude that custom tests will yield drastically different results from preset tests, and custom IBT AVX runs are not necessarily comparable.