Okay, let's be clear on what we're going for. As I see it, we want as granular results as possible, using a single roll, a single, elegant (hopefully even fairly simple) mechanic, and the minimum possible number of reference tables consistent with the granularity we're going for. Please don't hesitate to correct and/or amend that. And yes, "the level of granularity we're going for" is going to be a compromise we settle for because anything more granular is not playable enough.
Armor degradation would be nice, but it's probably too fiddly for too little return.
Damage reduction according to various armors... if you try to differentiate by both the attack form and the target surface, it becomes an ever growing can of worms. There is just too much variation in what qualifies as "armor," and even more in what can qualify as "an attack form." To keep that variability to a manageable level, personally I would differentiate between armors, but not differentiate according to the attack types used against them. That costs you things like a realistic presentation of the weakness of chain armor against puncture attacks, but in practical terms I don't think it can be helped.
It might be possible to make that weakness in the mechanics less obvious by differentiating armors not only by hits of damage absorbed, but perhaps a round of stun cancelled, 1hit/rd in bleeding reduction, reduced penalty imposed by the crit (-10 to all action vs. -15), etc. The rationale for what benefit to assign to what type of armor needs to be carefully thought out. It makes a certain amount of sense for plate to reduce bleeder results, as reflecting the difficulty of getting to those areas that bleed profusely in the first place, and the likelihood therefore of getting near it instead of on it in the second. A nicked femoral artery is serious, but it's still orders of magnitude less serious than a severed one. With chain being nearly the same from an archer's point of view as no armor at all, it wouldn't make sense to give a bleeder reduction there.
Any of this might work. I'd want to see various ideas along those lines playtested before I committed myself beyond "might," though.