My results are not as good as Lostfaith's 940 Black Edition, but for a cheaper cpu they look pretty good.
For all the results, I ran 2~8 hours of OCCT (or 8 hours of Prime 95) at 0.0125 ~ 0.0250 volts under the ones I marked there. So they are rock-stable, 24/7 every day undervolts/overclocks for this cpu. Aparently the minimum voltage that my motherboard supports is 0.8v (but I didn't tested it, I don't know why), so I didn't use anything lower. At this voltage I was runing even 1200mhz (but didn't ran prime in it).
The voltage curve is linear in all clocks, except above 3ghz. I didn't tried any higher OCs, as the step-up to 3.1Ghz already took that much more volts. The temperature was also in the 51°C, so I stopped. Probably I would have maxed out at 3.2~3.3Ghz. The motherboard was capable of more than 260mhz even with the onboard video enabled, but I would also rather not push it much.
The CPU/NB stock seems to be 2Ghz@1.275v, and I overclocked it to 2.3GHz @ 1.200V, and I'm runing it now 2.07Ghz @ 1.150v. Actualy, I was able to boot to windows and see HD videos @ 1.050v, but the OCCT would hang in less than 1 minute even with small data set even at 1.1v. Then ran some OCCT at 1.125v, 20min in each dataset size, and it passed. So I'm with 1.150v now (and I'm lazy to see if I can lower it one or two notchs). The HT also worked at up to 2.3Ghz, but I lowered it for some additional stability anyway.
My specs:
Processor: Phenon II X4 810 (stock: 2.6Ghz@1.3v in this motherboard at least) - serial number: "CACZCAC 0903 BPMW"
Motherboard: Asus M4A78T-E - AM3
Video Card: Onboard 790GX with sideport + 256mb UMA
Memory: 2x2GB OCZ3G1332G runing at 1226@7-7-7-1T 1.6v ("stock" is 1333@9-9-9 1.9v)
CPU cooler: Coolermaster Hyper TX2 with the pre-applied thermal paste - Set on silent profile on MB bios.
Case: Cooler Master Elite 335 - Fan set on silent profile on MB bios.
PSU: Corsair 400CX (80plus)
I also made an very rought estimate of my CPU and system power consumption at every undervolt setting. I don't have any wattage mesurement equipament, so I used
this PSU calculator and tried to simulate my system encoding a video. I picked the Intel C2Q Q9300, as there was no 45nm PII avaiable, and the wattages seemed reasonable.
In the graphics below, the red bar is the CPU power consumption alone, and the blue bar is how much the entire system is sucking from the wall, with the PSU inefficiency curve already included. In the thumnailed graphic I simply took off the system power, for a better view of the cpu consumption alone.

The conclusion is that, even thought it is possible to reduce alot the CPU power consumption undervolting it, I'm not doing anything to the rest of the system. Thus, my encoding, for example, will drawn less power overal with the cpu runing at 2.6Ghz than 1Ghz, as it will finish sooner. (that is actually a lie, because I always try to encode at 0,7 fps, regardless of cpu speed, but anyway)
Also, I don't gain much going from 1Ghz to 800mhz if I mantain the same voltage. In fact, it seems like 1~2W difference. So I'm using 1Ghz in hope that the computer needs to jump p-stats less often that way.
I'm now runing with the following k10stat configuration:
2990mhz - 1.2625 v ("3ghz")
2300mhz - 1.1000 v
1610mhz - 0.9375 v
1035mhz - 0.8000 v
I had some problems of hiccups in HD video playback, letting windows adjust the clocks, or runing in the unganted mode, thus I set k10stat to start with -ClkCtrl:2 and it was aparently solved. No problems so far. The complete configuration:
Additional images of the system at load and after some time at idle (room temp was around 25ºC). Shows the k10stat old bus speed bug (from v1.35 ajustable by -HtRefClk) , CPU-Z, AMD overdrive main screen, prime 95 and Speedfan.

EDIT: Updated K10stat setings