I agree with Mike here (perhaps a little less abrasively...)
Making decisions on a consensus on a unit statistic other than through statistical means is sub optimal, 100 people saying 2+2=5 still doesn't make it so....
I conducted a little experiment this morning, I rolled some dice.....
I rolled in groups of 4 to try and keep experimental conditions consistent, and I alternated collecting data to attempt to apply any external influence evenly
I rolled 8 dice needing 4+ to hit and 12 dice needing 5+ to hit, I did this 60 times each (I was going for 100, but the lure of a bacon sandwich and cup of tea halted that)
I recorded each total and took the average, the set of 12 dice needing 5+ gave an average roll of 4.133 hits, the set of 8 needing 4+ gave an average of 3.583 hits
this discrepancy can probably be attributed to a combination of human error in recording the data as well as punching the numbers into my calculator....
the reason 3x5+ is simply better however, is in the non-average results, a result of 7 hits came up 5 times more often in the 3x5+ group as it did in the 2x4+ group, we also had a single result of 9 hits in the 3x5+ group which is completely impossible in the 2x4+ group, there were also twice as many results of 6 in the 3x5+ group as there were in the 2x4+ group
on the flipside, there were 5 times as many instances of a single hit being scored in the 2x4+ group as there were in the 3x5+ group
hope this helps