And yet we still
subtract damage from hit points...

Frankly, this caused (and still causes!) more mistakes in the maths of D&D than having decreasing AC ever did IME. The study you mentioned involved 2-digit sums and differences and is much more appropriate in demonstrating how damage should be tracked cumulatively instead of being subtracted from a maximum value:
View attachment 383575
I mean, these are NOT the type of numbers used in AC caluclations in AD&D or d20 systems, so I don't find your conclusion quite as supportive as you do.
At any rate, I always find such discussions odd because I never recall anyone when I was an adolescent through college having issues with the concepts involve with the AD&D "to hit" tables nor the use of THAC0.
It was never IME or IMO a maths issue, it was understanding that +1 armor was one "class" (which is why it is called Armor
CLASS) better than without the +1. You moved "up" on the to hit tables, showing you had better protection and the attacker has to roll "one higher" to hit you than they would have needed had you not been wearing the +1 armor. Even when AC's were negative values, it wasn't a problem really.
THAC0 was simple enough. You recorded one number and subtracted the AC value of the target to find the number you needed to hit. You rolled, added any bonuses, and hit if your roll was at or above that to hit number, you hit. Alternatively, and what many people did, was subtract their attack bonus from THAC0, so once they also subtracted AC, you didn't add anything to the roll. Worked either way.
In general, the numbers we are discussing are small enough it was very rare, if ever, an issue IME. However, I very much understand why with 3E they need to change things! Bonuses and ACs became so extreme without the to hit tables that the maths actually would be vastly easier with an additive system. In fact, there were more errors in 3E in attacks and AC due to the crazy numbers you needed to track than I ever saw in AD&D.
But hey, that was just
my experience.