There is a huge difference between BECMI and 4e and DungeonWorld and 13th Age. BECMI and 4e are both games about killing monsters and taking their stuff. DungeonWorld and 13th Age are much more recursive games about being/playing some idealised version of D&D.
Which is about killing monsters and taking their stuff. I don't want to get into edition/game engine arguments though, at least at that level where people tend to have emotional commitments and loyalty to one system or the other.
Mine are also practical; my objection to subsystems is a simple one. Any time spent looking rules up in a rulebook or questioning these weird rules is time where immersion and flow are broken, and that's a steep cost to me.
I don't deny that there is a tradeoff. Resolution time/complexity is steep price to pay. The other issue is the learning curve for the GM. The more rules you have, the longer it takes the DM to get comfortable. But I would argue that the price on the learning curve to subsystems isn't as steep as you think, because it neglects the cost the GM has in learning how to ad hoc functional solutions to problems that come up routinely when you don't have rules guidance. That cost is just as and maybe even steeper than having more rules. This is because being a good rule smith is a skill, and not everyone has it, and certainly few or no novice GMs are going to have it.
Therefore to justify themselves subsystems need to be used extremely regularly (e.g. damage mechanics) - if it doesn't come up more than one session in two it doesn't belong.
That speaks to precisely the core of my problem with not having subsystems. Imagine you are building a new engine for a typical fantasy RPing game where player characters may be challenged with problems like evasion, stealth, skillfulness, diplomacy, magic and combat. Typically you see a design like this:
a) Evasion: Give a yes/no answer to the question, "Did he get away?"
b) Stealth: Give a yes/no answer to the question, "Did he go unnoticed?"
c) Diplomacy: Give a yes/no answer to the question, "Did he get what he we wanted?"
d) Skillfulness: Give a yes/no answer to whether he can passively hurdle some challenge - "Repair sinking boat, yes/no?", "Find clue, yes/no?", "Translate ancient rune, yes/no?", "Find safe shelter in wilderness, yes/no?" etc.
e) Combat: Provide a system for combat that allows for tactical motion, weapon use variation, armor use variation, dodgy targets versus relatively invulnerable ones, compare defensive tactics to offensive tactics, allow for granularity of wounds, the use of various stunt and maneuvers in combat, differences in combat with creatures of different sizes and shapes, fatigue, inflicting statuses, pain, shock, blood loss, etc., etc., etc.
f) Magic: Provide a robust system for situational breaking of the rules of the game and the imagined reality.
So you put that system in to play, and completely unsurprisingly players invest in combat abilities, see combat as a more reliable solution, see combat as a more rewarding and more enjoyable solution and non-combat scenarios only come up extremely irregularly and are not the focus of play. You'll also not unsurprisingly find that being the guy that does magic is a lot more rewarding than the guy that tries to rely on the gated 'yes/no' subsystem, and the DM tends to find the gated 'yes/no' subsystems so broken (or so evades the good fun stuff) that he weights the system in some fashion to always give a 'no', or that the designers ended up doing the same thing.
Alternatively, you see systems where the designers do something like this:
a) Combat: We've built an elegant system for combat resolution that allows us to compare defensive tactics to offensive tactics, allow for granularity of wounds, the use of various stunt and maneuvers in combat, differences in combat with creatures of different sizes and shapes, inflicting statuses, etc., etc., etc.
b) Evasion: We'll leverage the combat system to imagine running away as a form of combat.
c) Diplomacy: We'll leverage the combat system to imagine negotiation as a form of combat.
d) Stealth: We'll leverage the combat system to imagine hiding as a form of combat.
e) Skillfulness: We'll leverage the combat system to imagine a skill challenge as a form of combat.
f) Magic: We'll leverage the combat system to imagine magic as a form of skillfulness.
This produces a really fun rules system to read. It's really elegant on paper. It's really unified. It's typically filled with little examples of the system working. And it lasts about 30 minutes of real play before you realize all the little examples of the system working were very carefully chosen (perhaps without realizing it) to illustrate happy situations where the system works, and that in the real world it's just not so pretty and doesn't offer any answers and often little in the way of compelling resolution. The problem I have encountered is that the fundamentally important ideas in each different sort of situation that can come up in play are not abstract, but pertain to concrete positioning within the fictional space. In combat for example, such things might be, "Is the target flanked or not?", or "Can I grab hold of the target and hang on, and if I do, what happens?" The combat engine in order for it to be compelling has to take those concrete attempts to interact with the fiction and turn them into resolutions that fit the situation. But if the subsystem isn't built for that, what you end up with is nothing but abstract handling of abstract interactions. And that means very quickly, the players stop interacting with the fiction and start interacting with the rules. The worst offender for me is attempts to make social interaction work like combat, where the player learns to stop roleplay and starts declaring rules actions. The thing being simulated starts less and less being represented by the simulation. Previously, we might have actual conversations and dialogue at the table. Now we have, "I make a savior-faire check to implement my cutting remark maneuver... a 17, that should surely overcome his indifference, yes?... I inflict 8 ego damage, and his minions have to make a loyalty check." There are of course ways to work around those problems and things you can do, but they become ways of handling the system and thinking about the system and in effect 'house rules' to cope with system limitations.
Perhaps an even more clear example is handling 'jump' - something that involves a granular range of results - with a linear pass fail or linear resolution. D&D in 3.Xe RAW is really bad about this, returning totally dysfunctional results about how far a person can jump, because the size of the random modifier (1-20) is much much larger than the typical jump modifier. With no way to expect to know how far he can jump, and the consequence of 'fail' on the 'yes/no' gate being so high, no player realistically tries to jump the chasm. It's much easier to magically fly, and this is primarily a consequence of shoe-horning jump to the uniform D20 mechanic that determines 'yes/no' across the board! Elegance is the enemy here of practicality.
And it doesn't come up because we've more or less (inadvertently) deprecated it in the name of elegance. And that justifies to a large extent athletic ability also not mattering and not being worth 'wasting' resources on.