dtgreene: The problem you have here seems to not be with the stats themselves, but rather having strict stat requirements. Using the number in a calculation (rather than a strict comparison) can solve this problem, and having damage be slightly random can prevent one point of STR from being the difference between 1 or 2 hit kills.
Stats in games are used in combat, as input for formulas, and out of combat, for binary results. The problem you solve here is the combat output being a non-continuous function.
(Noncombat binary success/failure can also be randomized, but even in many roguelikes the outcomes are so important they aren't left to chance because the swinginess would depreciate tactical skill; a classic RPG player can just reload).
But the other problem is character optimization (with regards to combat effectiveness). It can be as continuous as the underlying math engine allows, but the number of maxima is finite and low. Maxing out the INT of a single-class wizard whose spells all work off INT isn't a choice. The assumption goes that players are trying to win the game, and the target audience for a combat encouter is going to be these maxed-out characters: "the boss encouter of dungeon 3 is a green dragon, so we expect the party to have Attack 25, either Dodge 14 or elemental damage reduction, and 2nd-tier spells". Leaving the option to not max out serves no purpose.
dtgreene: INT does have a purpose here: It can be used to differentiate a pure mage (who would get high INT) from a hybrid (who would get lower INT making their spells a bit weaker, but would not need to rely on spells). Also, it makes it easier to tell what effects equipment and passive skills have on the character's spellcasting ability (assuming such equipment and spells have visible effects on the stat).
(Note that, here, I am speaking from a JRPG-esque perspective, where you usually have stats but don't typically control their growth.)
The end result of designing a combat system in an RPG is a set of available actions with their associated resource costs and chances of success; a character generation system adds restrictions on which subsets of actions are available on individual characters. When you have this math construct and made sure it works, it doesn't need to be retroactively justified by making the ideal output a fake function of some other numbers, and especially it doesn't need to be broken by adding more moving parts.
Fighter-mages as multi-attribute dependency (MAD) builds have historically not worked well (or, indeed, at all). MAD breaks the system's benchmarks.
- Absent combos between things humanity was not meant to combo, a MAD character is dead weight.
- If combos exist, there's absolutely no guarantee the character would be in line with a single-class wizard or single-class warrior (while a sliding scale of physical attack stat vs magical attack stat isn't what causes the problem -- multi-"classing" does -- it certainly doesn't help). If, say, there's a "wizard" INT spell that delivers a nasty touch (requiring a squishy wizard to be next to an enemy and hit with a 1/round attack) and you get that spell on an archer precision-shooting 20 arrows per round at targets up to 3 miles in range, it's a problem, and one that might not be isolated and fixed beforehand because the amount of possible combos is some crazy combinatorial number with a Hungarian surname.
So if you want a spellsword, it should just be a separate class. If you feel the advantage of having a choice between a magical and a physical attack on a single character needs to be accounted for, you can just make a spellsword's attacks less powerful by changing numbers as you see fit, by amounts that you can adjust independently, without wrecking the foundations of the system.
As for equipment bonuses, they can just modify the important characteristics directly, from scaling effects across the board to spot-modifying specific abilities. So a ring of INT +2 would be a ring of power +2 that adds two levels to pell effect calculations. It also means it can't be easily given to a rogue to increase disarm traps, but exactly because stat increase items need to play it safe and provide "balanced" gains in the best use case, they have never been much fun. (Example: suppose paladin healing works off INT, which is usually low because they don't have much else to do with INT. A massive INT bonus item that'd allow a paladin to drop a huge meaningful party heal is nice, but it'd also allow a wizard to nuke each and every encounter. On the other hand, what's a meaningful overcap bonus to a wizard would be a waste of slot for a paladin because a better use of an action would be still "bash a sinner's face in with a warhammer and pray for a crit".)
dtgreene: What about "learn by doing" linear games? (SaGa 2. when played with a party of mostly Humans/Espers, would fit; if played with a party of Robots, stats become equipment dependent, with many different equipment options (like the ability to wear multiple suits of armor for more defense).)
Learn by doing linear games are either optimization nightmares (if they don't allow grinding) or purposefully forgo target difficulty as a design tenet (if they do) and I'm not going there.
