T. Foster said:
I agree with this completely. Gygax went with a d20-based system (rather than the previous 2d6 based system) for combat and saving throws in D&D because he wanted a flat probability distribution and (I suspect, though he's never confirmed this) heavy random element, but retained d6 and d% based rolls in most other areas. The move towards using d20 for everything for the sake of consistency is, IMO, a mistake. (And, FWIW, one of my long-standing house-rule projects is an attempt to move D&D back to 2d6-based rolls for combat and saving throws -- but that's a tough battle against a very ingrained tradition.)
Err - if I might note, Gygax uses a flat probability system with d6 and d% as well. The grain of the system changes, but it's still a flat probability system. This is as compared to 3d6, which gives a curve, and is what the poster wants.
In all of its incarnations, D&D has used a flat probability system for resolution. (The exception is opposed rolls in 3e, which give a "curve"... well, a triangle.

)
The *advantage* of a flat system is that it allows a +1 bonus to always have the same meaning, and for a wider range of values to be significant. It also allows the someone who is worse to compete. The *disadvantage* is that someone that is better doesn't have as big an advantage.
Personally, I think a curve system is at its best when you want the players to succeed most of the time, and the values are adjusted in such a manner. Compare
Runebound 1st edition (which uses d20) to
Runebound 2nd edition (which uses 2d10). I think the 2e version of combat is superior because it's much easier to gauge whether you'll succeed or not.
However, for a game with as many power levels as D&D has (and I speak of all editions), then the flat probability system is better.
(One can see one of the big flaws of 3d6 in the GURPS 3e skill system, where skills were based on an attribute, normally Int. If you got a 16 Int, your chance of success was almost total for everything...)
Cheers!