The amount of benefit you gain from a reroll though is directly proportional to the amount of variance you expect in the first place. A reroll on a D4 moves the expected average from 2.5 to 3.13, a gain of .63. A reroll on a d8 moves the expected average from 4.5 to 5.81, a gain of 1.31. A reroll on a D20 moves it from 10.5 to 13.82, a gain of 3.32. It's actually better than that though - since your d20 determines what the *outcome* is - as you've said, a miss = 0 damage. Consistently hitting rather than missing has a *huge* impact. It also determines when you crit - critting 27% of the time instead of 15% of the time, just by leveraging the mechanics of the game and using advantage, is a large benefit.
Meanwhile though the damage reroll only determines the outcome if the additional damage kills them while otherwise they would not have died. Otherwise they take damage either way, it's just a matter of how much, so the benefits are far more incremental. Furthermore the benefits decrease as the size of the die pool increases, so your proportional benefit dies off as you increase the total damage figures, whether adding additional static modifiers or more dice.
The point being that you can have more damage *and* more accuracy, by combining advantage (for accuracy) with ways to leverage that more favorable probability curve into small decreases in some of the gained accuracy for a large damage bonus. If you need a 5 to hit, 80%, and you switch to having advantage, 96% to hit, you can take a -5 to hit *and still be at the 80% you started with*.
As far as RNG being "shit" or providing results other than the expected outcomes based on large sample sizes, I have not observed any evidence of this. Do me a favor, cast firebolt 100x at targets where it should do 2-20. Look in the combat log, and record each damage amount. Toss the results back, and we'll see what the average is. If it's 2.5 instead of 10.5 as expected, I'll do the same. If there's a legitimate issue with RNG let's find it and prove it so Larian can fix it. I've gotten results consistent with expectations though - if attacking at 65%, I'll hit twice for every time I miss. When using Firebolt I'll get an 8 on the die as often as I get a 2. But hey why take my word for it? Test it yourself, let us know how it goes.
