Well well.. we're making good progress.

See the above 2.6G 20hr stable?
I run that setting daily... replicated it just now at 30C, 35C, 40C, 45C, 50C and 55C BIOS (70% load)/EVEREST -> results:

30C idle/load -> stable
35C idle/load -> stable
40C idle/load -> stable
45C idle -> shaky
45C load -> stable
50C idle > no boot!
50C load -> shaky
52C idle -> no boot!
52C load -> mostly errors
55C idle -> no boot!
55C load -> full of errors!

Shaky = Errorful on and off but altogether unstable (freezes/hard locks)

We have established a temperature bottleneck with Phenom guys. Achim (justapost) experienced something very similar and I've never had high temp readings to be able to tell but today my temps were simulated as high as his were before and what he found is exactly what I'm finding...

Freezing/hard locks are known as a very possible temp issue on any setup.

I could not measure the IHS temp nor did I have probes with me but I used the normal method made for us to use: software monitoring tools. The temperatures reported were the above with software/BIOS.

However, I do not think that was the real temperature. To me it looks highly likely MSI board/Phenom is reading 10-15C below the actual core temps as we know and there is no CPU I've seen which fails POST 20C below max danger temp limit. In Phenoms case, that's 70C, anything below that (real) should run fine. Those temps of 50C must have been 60-65C real IMHO.

Jack (JumpingJack) briefly mentioned relatively high stock temps with his Phenom sample too and that I couldn't corroborate with Gigabyte/MSI/DFI+Phenom testings but I kept checked around and found two other ASUS+Phenom pairings and they both idled/load pretty high up according to software at same MHz/Volts as us, like him. Thus, to me it seems highly likely that the ASUS boards have the correct temp. calibration for Phenoms and that the MSI boards show 10-15C below actual. Otherwise, 50C will never fail POST and freeze whilst in the BIOS.

One other thing I tested but could not measure for lack of equipment with me was MOS/inductor cooling: yes you do need it or you'll fail stability testing at stable MHz easy. Especially at load, typically lower ambient/PWM/CPU temp will get good stability it seems. Guess I better get my skating shoes on.

Gack I was certainly surprised.