Your example exemplifies the whole issue with the prevalent incompetence. The fact that an easy task will be failed 40% of the time means that PCs even at levels where they should be competent fail to be reliably so.
Without total randomness working in their favor, PCs simply aren’t very good at doing stuff. No way should PCs be failing easy tasks at 5-6th level a third of the time. It shouldn’t require stacking talents and having multiple training packages for a 5th level PC to be consistently good at stuff that fits into the easy category.
This is an old design flaw that’s remained in the game for years. It’s exacerbated if the dice rolls have a true bell curve.
Let's put aside the dice manipulations discussed in this thread for a moment. The details are going to vary between editions but in RMU, a 5th level character can have up to 10 ranks in a skill (a couple more for cultural skills), so that's +50, plus a stat bonus, plus a professional bonus if relevant. For a skill they are not training intensely, have no cultural background in, have no talents for, not a professional skill, they might have 5 ranks for +25, so perhaps +30. Easy is +20, so +50 on the roll. If they can't find a +5 complementary skill bonus they aren't trying, so +55. If the skill is something they can make gradual progress on (a percentage maneuver), they are almost assured success, it's just a question of how long it takes. If partial success is possible, the chance of actual failure is only 20%. It's only unforgiving maneuvers that are pass/fail that they will have the higher 45% chance of failure for. The character with 2 ranks/level, if it's not a professional skill, will have maybe +55, +20 for Easy, maybe +5 complementary skill bonus, so +80. If partial failure is possible, they will only truly fail on an open-ended down roll. Even if it isn't, the chance of failure is only 20%. A professional skill could have another +10 and the stat bonuses will likely be higher, so it's quite possible they will only fall short of Success on an open-ended down roll.
You mention training packages. If you are looking at RMSS, you are starting with your culture ranks, and you have your professional bonuses up front. Plus your stats will be very close to their potentials by 5th level. Bonuses will generally be higher. Near success on 91 will usually let you succeed in one additional round. 5th level RMSS characters are very competent.
I may be a little confused regarding changing to d20, since a d20 doesn’t yield any different distribution than a d100 except being less granular. It’d require something like 2d50, 3d33 or 5d20 to yield any kind of curve - unless I’m missing something.
The OP proposed replacing the tens die with a d20 numbered unevenly, e.g. 00,10,20,20,30,30,30,40,40,40,50,50,50,60,60,60,70,70,80,90. So that increases the chance of results in the 30-79 range, and decreases the chance of results in the 1-19 and 80-100 range. It's not a very smooth bell curve and the impact is much smaller than rolling three dice. I posted above how to plot that distribution on Anydice. That's what I was referring to.