I would like to point out the downside to this discussion, but I would first like to say that I liked the idea and I did an analysis of this a while ago to see how it worked out.
Definitely the idea can have downsides.
For combat: Criticals are determined by rolling a max on the die roll. If you start to decrease the dice too much, you end up with a much higher chance of criticals and at higher levels, they can be unbalancing/brutal.
I'll note that my recommendation was that a 20, and 24 only are crits (and a 30 if you go up that high). Otherwise, I suggest you can't get a crit. This actually leads to slightly more, but not unreasonably more crits if you get the really big dice.
Outside of combat: For each +1 you need to drop the die by two die slots to keep the same progression. (i.e. d20->d18->d16->d14->d12->d10->d8)... so a -6 is really a d8 roll and not a d14. That keeps the same average chances of success for the mid-line numbers (i.e. average goes from 10.5 to 4.5). The problem with using that progression is that rolling with rolling a d8 and adding bonuses you will ALMOST never achieve a result of 15 or above.
Where if you roll a d20-6+bonus, you still have a good chance to make a DC of 15 or higher.
A few points:
I don't believe there is a d18 in the progression, which really makes the smallest shift effectively an average -2 from a d20. That is part of the reason I left a +/- 1 in the system.
I also modified the shifts in combat table with your statements in mind. I think that in general you shouldn't have more than a one die shift up, or two die shift down for most checks. That is why I suggested that only the largest modifier should apply.
When designing the modifiers you need to keep in mind that both the average, and the highest possible result are changed. For example, I think that rolling a d14 while in banded mail or a d12 while in plate mail instead of a d20 is an appropriate change given that the DC will usually be 10 or 15. d12+bonus is definitely a different curve than d20+bonus-8, but for a situation with a +1 ability bonus a 15 is impossible in either case and a 10 is possible 33% of time with d12+1 and 20% with d20-8+1. I'm okay with that probability shift, but agree it is a change.
Remember shifting dice down causes a bell curve change on the probabilities of success and makes a major change to the chances of success. Where leaving the dice at a d20 keeps the + and - modifiers to be a linear change the the probabilities.
Plus, the purpose of DCC RPG is not to make it more complex on the fly, but instead to simply use the funky dice. What does that mean?
All funky dice are pre-written on the character sheet and you can build a dice pool before your session. You quickly roll your pool and add any extra bonuses or penalties. No hunting for dice or extra math to calculate the dice required.
(Yes I know that he is using d20/d16/d14 for combat, but the combat system is a special use case right now because AC is not really a linear progression because tougher opponents do not ALWAYS have a higher AC. They can be represented by extra HP, etc).
I like the changing probability curve, and I don't think grabbing a different die because of a combat modifier is that complex. I think it would be fun to be able to say "I Charge" and grab the d24.
As much as anything I think this unifies the game systems because you do checks to hit and checks to take other actions exactly the same way.
However, I'll acknowledge some other drawbacks while I'm here:
Not giving crits when you are rolling dice less than a d20 makes taking a 1 step decrease on an attack die rather significant. Not sure I like the idea that cover prevents crits. I would probably suggest cover grant a change in AC in a system like this.
That still leaves the problem of no missile weapon crits at medium/long range. Whether that bothers you, or even would come up much in play, is probably a question of taste.
A system like this would make it harder for folks confused about the underlying math to quickly convert special case modifiers in non-DCC written adventures to an appropriate number of dice steps. (Although the rule of thumb of 1 step for the first +/-2 and a step for each +/- beyond that is close enough in my opinion.)
What it really comes down to is whether the design desire is to come up with a unified system that embraces the funky dice, or to come up with a unified system that has the funky dice sprinkled in occasionally for flavor, or to come up with different subsystems depending on what you are doing. Right now I think things are pretty solidly in the third option but moving at least towards the second.