Thanks for the comments, Mike L. I agree that our chip is probably not the greatest. Your result on water is significantly better than we saw on above ambient cooling (I only managed 3.8GHz at 1.55V). I spent quite a few hours with the rig, so I'm pretty confident that I didn't miss anything. It was really tough to find a 100% Prime95 stable clock speed and had to scale back quite a bit to retain 100% stability. I was able to get some pretty impressive clocks without worrying about Prime95 stability, but 100% stability painted a very different picture.

Hardware Canucks did a good Phenom II overclocking writeup as well and saw similar results on phase. Their highest CPU-Z validation was about 500MHz higher than their 100% stable 3DMark Vantage benchmark.