I agree about most posts on this subject, and this proposition (avg over the last 5 min) is the best I can think of.

Sure we want to point out the rigs that clearly underperform, but the current indicator is not better than a random flag!
Immediate speed is false since mining speed always fluctuates around a certain value, and we certainly don't want to advertise the low value of the range (for instance my rig is between 4.4 and 4.8, and the 20 minutes average is always around 4.68-4.7 so I advertise 4.6 even though I could advertise a little more, and I would definitely not want to advertise 4.4, that would be a 5% flat loss!).
So it has to be some average of the past. 5 minutes seems fair to me (10 might work as well - I believe we will actually see what is working best with some trial and error), but average on a longer period is not a good thing since the rig might have gone through fluctuating hash rates over the last day (by changing the rig for instance or even switching from one algo to another - scrypt vs X11 is a x3 factor).