that is immersion not game play.

Luring a strong MOB into a nest of another MOB so they kill each other and you slay the 'victor' isn't game play?

MOB on MOB interactions that players can figure out and use to their advantage would be a great improvement over the scream and charge killer mannekins (on rails) most of these games have. Normal patterns, logical behaviors/interactions, territories, valid/logical placement/group dynamics...
Enemies should runaway, run to defensible positions, ambush players, assist players, seek reinforcements, cringe and plead for their lives, ignore players(if big and bad enough), feign attacks, fight defensively, etc.. These different interactions ARE scriptable at the current server NPC processing loadings (offset largely on no longer having minefields of countless enemy entities across most of the world). Fewer more interesting computer run opponents (more challenge+reward and less repetition) with more suprises and more tactics to match the situation. The entities can actually look like they belong in their situation (no spawning out of thin air next to the player).

In MMORPGs you will want to see more instancing (and a large world to make the instances far apart) so that you can choreograph more interesting quests/missions/encounters without the interference of other players (who as we know will willfully interfere with other players to get their infantile jollys). The more complex something is, the easier it is to fall apart and external player interfence has to be exculded except for the players group.
Bigger player mobs can happen in other scenarios (and are interesting in their own way) but for most of the game disruption of the players experience needs to be eliminated. With a sufficient bubble size the player can decide which way to approach the situation even round about paths if allowed and dont have to be restricted to 'getting out of the way of the next boatload of newbies' and can take their time to figure things out.

Dynamic (generated) content more complicated than simply running into a MOB can be done including scenery props and terrain using hierarchical parameterized templates controlled by world spanning influence maps (with local recursion) to vary the details of the experience. Those would be controlled and change with larger world trends and patterns (which could be influenced by player activity). AT the highest level you probably would need some GM controlled tuning/adjustments to keep the whole system from falling apart (trying to write a program to do it trouble-free would be too much work)

You have a instance bubbles and build what you need in it for whatever the quest is (or even just random encounters) The bubble is on whatever part of the world it needs to be and its border can be seemless. Scripted events/dialog/explanations can be part of the larger template with numerous substitution possibilities/alternate factors. Quest givers/intermediate checkpoints/target areas can be generated and worked phase by phase. These themselves are triggered by the larger game plot lines that give the whole system some cohesion as the 'plot' (itself maybe a tree with different paths) . They are tailored to the player.players group and can be regenerated whenever needed (for player playtime flexibility)

You can stay away from any (generalized) Real AI as the local situations are largely choreographed/balanced (and can use conventional FSM scripting methods) and the high level control of the gameflow can just be cellular automaton type behavior (high level entities that guide/flavor the local situations ). Its more varying behavior from the computer controlled entities and the behavior scripts themselves can be heavily template based to improve production efficency and allow tools to streamline creation/debugging/testing.

Why dont games (game companies) do this (and probably wont for a long time) ???

The scripting is about a magnitude more complex (random substitutions that need balance trimming and illogical endcase elimination)
Templates can greatly streamline creation but there will simply be more factors involved that need to be planned for 'situations' and how they fit into the larger game. From what Ive seen alot of the game companies have enough trouble making their simple static scripting work correctly.
Old QA rule - twice as complicated, four times the debugging and game companies think nothing about skimping on QA if they can get away with it.

Players would have to actually learn how to think and pay attention to clues (even when you lay them on thick and they have logs to go back to refresh their memories - might actually need dialog scene replay which might be neat). The game mechanics will have to support a wider range of effects and choices/outcomes of situations (versus the boring fail=die, win=loot) which actually in a more complex world would shape the players future world experience with the consequences(and not just these static paths and tech trees of current games)

You cant so easily have an online wiki tell you exactly what to expect and do for every bit of the game experience (there can be enough combinatorics of situation variations that you wouldnt be likely to experience the same things anyone else did most of the time).

WIll casual players want a more detailed game or just reject anything that isnt the mindless conditioned monkey with the lever and the m&m rewards the current games largely provide?

Later, with better tools, niche games (with much lower player counts) may be a whole lot easier to produce (thing the computer publishing revolution) and we could get this different kind of game. .