The issue with hit points is that they are very close to the sequence of coin tosses. When my guy hits your guy and your guy loses 5 hp, we don't know: whether my attack was strong, or your defence poor, or a bit of both; we don't know where I've hit you, nor what the nature of the injury is (although perhaps we can infer it's minor, from the fact that your performance is not debilitated in ay way); although, at the table, we may have spent a minute or more resolving the attack, and will have to repeat the procedure several times before the outcome of the combat is known, we really have no insight into what is happening in the fiction except a general sense of which way the tide of victory is flowing from round to round.
It's not quite as divorced from the fiction as the sequence of coin tosses: for instance, STR and weapon properties factor in to the attack; armour worn and DEX factor into the defence. But even there, we can't tell - when an attack is actually made and hits - whether it was the quality of the attacker, or the deficiencies of the defender, which resulted in the hit. If we want a richer fiction, during the course of the fight, than simply "It's going A's way; woah, B just made a big comeback and now it's going B's way", that all has to be injected.
(Contrast, say, RQ with its attack and parry rolls, or Bloody Versus with its attack and defence pools. It's true that RM or RQ doesn't give us as much detail as a serious student of martial arts might want to know, but it at least picks out the difference between an Errol Flynn -style swashbuckling duel and a Tarantino-style mutual gorefest. At the level of the fighters, it gives us descriptions of the fiction "He dodged!" or "He failed to dodge and got skewered through the thigh", though it doesn't give us descriptions at the level of their individual body parts or weapons. This relates back to what I said about the wargame example: part of judging something as sim is working out what level of description is salient. For a wargame it is units, logistics, commanders etc but not individual soldiers; for a typical RPG it is individual fighters, but most players don't worry about those fencing details that you call out.
Pemertons post is pretty much spot on, and I hope he's getting across what we mean when we keep using the word granularity. It means how far in or out you are dialing the scope.
Every atom in your body has it own individual thermal energy. You could, if you have a StarTrek holodeck level computer model your body temerature by calculating every single entropic interaction. Every brownian motion, every electron jump and photon release. But it would be crazy and useless. It's dialed too far in to even track metabolic processes generating heat like digestion and muscle movement. Dial out to the molecular level you can at least get some kind of relatable information about how your sweat is cooling you by evaporation or your ATP cycle is incresing your body heat. But it's still billions of interactions per second. So you don't track it at that level. In fact, without artificial instruments your ability to resolve your own body heat is limited to feeling hot or cold. And your response is limited to gross systemic actions like taking off a sweater or eating hot soup. Thereby rendeing the molecular detail level utterly superflous.
In an RPG you generally don't care at all if your character is chilly or sweaty unless you're in a situation where it becomes important like an extreme climate. And then it's usually resolved after the fact with something akin to a con check. Becuase no one really cares to closely examine the fiction of whether their fighter kept cool by loosening his chestplate or fanning himself with a stack of hobgoblin scalps.
In real life, if someone is shot in the chest and drops dead you might say "Aha! I know exactly what happened!" But you would be wrong. You don't know if he died from shock, blood loss, or some other effect. You don't know why the guy next to him who took an apparently identical wound didn't die. My answer? One guy rolled more damage.
If you spend any amount of time studying the injuries people do and don't die from you'll realize it's impossible to model accurately. Any system, no matter how pretentious, is never going to actually "sim" things out in a way that models reality. Likewise you might cry that hit locations and death spiral systems are more accurate than D&D. And they are, for some people. For others the adrenaline response is strong enough that they might not even know they are injured until after the fight. I had a friend who once told me "The first time I was shot I knew I was dying. By the seventh time it just made me mad." Sounds like leveling up to me.
If you are playing a mass battle you do not model every individual parry and thrust of a sword, you don't even model individual wounds. You generally know if a unit is alive or dead. In some systems a unit might have more than one wound. In others you don't even track people on the individual level. When you watch a war movie and the hero is trudging through the aftermath of the battle searching for his buddy, are you upset that you don't know if the guy who killed him won by being stong and tough or quick and nimble, or do you just empathise with the hero when he finds his dead friend?
Recall that D&D grew out of war games. Gaining hit points as you level originally modeled the fact that commander units would be capable of survivng more than one hit in a mass battle. Hit Points.
Then people wondered exactly how that works and Gary (bless him) mumbled something about how they weren't literal and ignored how things like blowgun darts and falling damage let you paint the model into a corner. Some people decided that they didn't like the disconnect between the fiction and the model and came up with systems that modeled (simed) a process that better matched their own ideas of how things should work. Other didn't like the disconnect between the fiction and the system and came up with better fiction. I fall into the latter camp. But neither one is wrong, its just different approaches to solving the same problem.
In both cases it isn't really the level or accuracy or the granularity ofthe sim thats the problem, it's the mismatch between what the system portrays and what the fiction portrays.