Yesterday, I plotted some curves comparing let's say 1d20, 2d10, 3d6+2 and 4d5 (add the total, add modifier like d20). Basically, the probability to have around 11 is higher as you take more dice. The probability to get a "critical" failure is as low as having a critical sucess. I love that. Now between all these methods, you just have to choose how probable getting a critical result is. I'd like to test each of these rolling methods to see how it works out in practice. I have the feeling that this would be better with defense bonus per level (to keep the net bab about 5-10 lower than the net AC) and much fewer hit points. Hitting would take longer but once you hit, you kill. Opposed rolls might work well too.
The rolls would be much more consistent (as if you were not rolling dice at all) but with some fluctuations now and then.