Single player games, and most other multiplayer games, are quite deterministic in regards to how many entities can be in your area of interest and needing physics stepped, while on the other hand when you are looking at an MMO type game it is generally non deterministic as to how many actors can be in your area of interest... maybe 1, maybe 5, maybe 50, maybe 100.
So even though the cloth simulation would be run on the client side as purely cosmetic, you now have to account for how many instances are you going to run/step and if you are outside of that what are you going to do? If you only process cloth physics for N of the closest and replace the others with some type of static mesh or anim how is that going to look? Also what is the transition going to look like if the N closest are constantly changing (i.e. a large group meeting) and you have to swap between cloth physics and whatever your stand-in technique is... or are you just going to grind out processing of all entities regardless of the milliseconds needed? So it won't always be a technical decision either as it may technically work properly but on the flip side it could look really bad and/or hamper performance/frame rate therefore the game is better off without it.
Regarding physics that actually matter to gameplay that is really a separate topic in comparison to "fluff" physics.