For being an English geek and horrible at math, I do loves me some probabilities.
On the whole, bell-curved systems are pretty cool ... just so happens that my favorite system uses a single-die system, which is okay with me too. I LIKE the curve probability, but I can handle the d20 (though it does start to break down pretty badly when bonuses exceed the variable). It actually sort of sucks at the low levels because of the huge variance ... so no happy medium.
For myself, I think what the OP is talking about is maybe more conceptual and mathematic. That a GM saying: "The DC for a hard task is 20, I call this at 22 as it's a little harder than usual but not much." is the same as a GM saying: "You've chosen the dice and the success, but I think this should require 4 instead of 3."
Which, yea, games-is-games. One can say that the d20 Player has all the control over how his character performs by crafting his story in such a way as he attempts Y action instead of Z action and the GM is bound to utilize the DCs found in the DMG/PHB and shouldn't cheat the players like the naughty bad man we all know he secretly longs to be.
(As far as the difference between curve and linear and fumbles ... I like the SpyCraft 2 approach where there's no guaranteed failures on a 1, but for most actions a 1 is an "error" and even if it's a success the GM has the option of spending an Action Point type thing to "activate" it. Mixes story and randomized numbers quite nicely. Sometimes, yea, maybe even a skilled person should fail 5% of the time for some wacky story reason ... sometimes he should only fail .0005% of the time while Bob the Bunglin' Redshirt fails 10% of the time outright. In part, the people at the table decide. ))
