If players want to use all the dice, I just run D&D.
Percentile system is of course Basic Roleplaying / Runequest. The purpose of other dice is to skew the probability curve slightly. d20 for example, is much more action oriented using a critical hit or critical fail mechanic, Literally a 1 in 20 chance of having a dramatic success or failure (Will usually happen about once a four hour session. D100 not so much crits and critical failures 01 and 100 respectively, happen about once a month if a group is playing weekly sessions. Using dice pools (like 3d6 for example) introduces bell curves, meaning that most things will be median, and the probability critical hits and failures drops dramatically to where you may only see one every six months or so, or even longer with more than three dice in the pool (1 in 216 chance of rolling three 6's for example... 1 in 1296 chance of rolling four sixes).
This is what I was getting at when I said people who analyze probabilities in RPG mechanics often confuse x for f(x). What you say is true for x. It's usually not true for f(x). In the examples you name, f(x) is always one of four possible outcomes--hit, miss, crit, and fumble. In each example, the crits and fumbles are the extremes and rarer. The hit/miss results are more moderate and far more likely. So the probability distribution for f(x) is as good a bell curve as one can hope for with only 4 data points. Notice even the mechanics that have linear distributions for x produce bell curve distributions for f(x).
I would also point out that it is not necessarily the case that multiple dice even make for rarer extremes in f(x)--there's no reason to assume, for instance, that only 1 and 100 are the crits and fumbles on the d100. Two of the percentile RPGs I play use one-tenth the base probability for success as the probability for critical (e.g. if your percentile rating is 60, you crit on 1-6, succeed on 7-60). The other percentile RPG I play generally has it at one-fifth. The 3E/d20 system had kind of a neat way of doing crits. It varied by weapon, but I think most often, if you rolled a 19-20, it was a potential crit, calling for a second successful hit roll to confirm the crit. This essentially made the odds of crit equal to 1/10 the odds of hitting at all. Same as the first two d100 systems I play.
So my position is that it's f(x) that really matters while x is largely irrelevant. And since the probability distribution of x does not
necessarily correlate to the probability distribution of f(x) (and in fact
does not in the vast majority of RPG mechanics I've seen), I consider probability analysis of game mechanics which invariably focus on x to be distractions and not very insightful.
Also, because I focus on f(x) rather than x, I have increased appreciation for the most standard plain Jane mechanics. Like D&D's roll d20 to hit, then roll funny shaped die for damage. Say a fighter needs a 13 to hit and does a d8 damage. When I look at the distribution of f(x,y) (x being the hit roll variable, y being the damage roll variable), I see there's a 60% chance of doing no damage, 5% chance of doing 1 damage, 5% chance of doing 2 damage, and so on. Much like the 4-data point "bell curve", this also creates a crude curve if you can forgive the limitations of the low resolution data set we're using. And that curve is more like a Pareto distribution--in this example, 83% of all damage the fighter deals will come from the best 25% of the fighter's attacks. Pretty close to 80/20. This satisfies my sense of naturalism far better than chasing bell curves, and strikes me as far more sophisticated as it keeps the game deceptively simple.