I found it helpful to make a spreadsheet grid, volts by Ghz, to track which combinations did or didn't work. One tends to see a line of feasible combinations, before one gets into crazy territory and voltages bend sharply upward. And 3.2 to 3.5 Ghz is not crazy territory.
For my G0 Q6600, the following pairs work:
3.2 Ghz, 1.28750 V (62 C max at full load)
3.3 Ghz, 1.31875 V (+0.03125)
3.4 Ghz, 1.35625 V (+0.03750)
3.5 Ghz, 1.40000 V (+0.04375)
So my numbers curve up a bit from a straight line, but I'm not yet at the brink.
One little-understood source of instability is "refresh to ACT", for which there's an advisory over at the OCZ support forum, and a mention in Kris Boughton's latest mobo review on AnandTech. I came back from a trip to find formerly stable overclocks now unstable, and started raising Vcore as a knee-jerk response, until it became clear that wasn't it. I then found a better 3.2 Ghz overclock 9x at 356 FSB, 1068 5-5-5-18 DRAM than the 8x, 400 FSB, 800 4-4-4-15 DRAM that was now failing me, and decided to live with it. But it was nagging at me that my mobo couldn't handle 400 FSB at tight memory timings.
Relaxing "refresh to ACT" to 54 bought me absolute stability. Nothing was wrong with my mobo or memory, just one more thing to learn, instead. Hey, it was getting boring until this issue, but I can see why people buy Dell.