The OP's friends are claiming 5.0 Ghz -- likely as a matter of bragging rights. I am skeptical about this. First -- air-cooling wouldn't likely be sufficient to handle the temperatures for such an overclock -- whether 2500K, 2600K -- even 2700K.
People often get their systems "tentatively stable" -- meaning it boots into Windows and allows for use of some desktop apps. then they run some stress-tests, perhaps 10 iterations of IBT "Standard" or "High" and call it a day. But that doesn't cut the mustard for "rock-stable 24/7."
It's my understanding that the last known published spec for an Intel "safe" voltage upper-boundary was 1.38V for Nehalem. [Correct me if this is inaccurate.] The Nehalems generally went into production using a 32nm lithography. Sandy Bridge also employs a 32nm lithography, therefore by a simple inference, it should have a similar safe upper limit of 1.38V.
I have a "pretty good" i7-2600K with superb air-cooling. I don't fiddle around with insufficient voltage settings, and I stress-test with LinX or IBT for 50 iterations. After it passes those stress-tests, I may add about 4 to 5mV in the voltage settings.
At 4.7 Ghz, my monitoring software under severe load shows a minimum actual voltage of between 1.355 and 1.36V. The unloaded "turbo" voltage at the end of a stress-test will typically register at about 1.37 to 1.38V.
I might TRY for 4.8, now that I have my temperatures pegging a maximum of 69.5C (4-core average maximum). But I'll have to up the voltage, and the recorded voltages are likely to be 1.37/1.38 severe load, and maybe 1.4V for the unloaded turbo reading at stress-test end.