The multiple d20 system is even worse (IMO) than the d30 system. I cite: Skill Checks and Ability Checks.
In RL, randomness is less of an element in certain factors than others.
Let us apply the 35 v 68 argument, albeit scaled down to -3 v 30. Now, let us take Knowledge (Mathematics). Using Crothian's number-crunching, we estimate that if Stephen Hawking (20odd ranks+2skill focus +7/8Int mod=+30) were to try to solve a problem at the same time as 100 profoundly mentally retarded individuals working separately, at least 1 should solve it quicker than he. Strange, no?
Or here's a better one: Grapple Checks. Imagine a frog (Tiny, Str 1) vs a Polar Bear (Huge, Str 27). The difference is (16size+13strength=29). Without number crunching, we can quickly see that the difference is less than 33 and hence there is a greater than 1.49% chance of the frog winning. If we lined up 100 frogs and they wrestled 100 polar bears, can you really imagine a single frog winning?
Randomness is all well and good. With regard to saves and attack rolls (where there really is a substantial random element) the auto-hit/auto-miss element takes care of extremes. The d20 system allows a fair amount of randomness to compensate from the big modifiers (a +10 in d20 is very strong, a +10 in 3d6 is virtually insurmountable) that DnD uses. d30 diminishes these modifiers by a factor of 1.5, and multiple d20s do so as well (though I'm not sure by what factor). If an alternate system is to be used, the entire game needs to be overhauled: simply shunting DCs and upping the dice is not good enough. By all means convert to d30, but it should be true d30, and not a d20/d30 hotchpotch.