You can paper over latency by immediately showing animations and stuff clientside, and rolling back if the server decides it didn't happen - most games do do this. But there's not a multiple second gap between e.g. you activating a spell in WoW and the outcome being registered by all players involved - that would be actually unplayable as you say. There's also sub second cast times combined with spells that interrupt/stun those other spells.
Nowadays there's not much difference between a lot of MMOs and what I assume you're referring to by action games - see New World[1], Black Desert, Tera, Elder Scrolls Online for a bunch of examples of games that have fast paced combat with small cooldowns and reticle aiming while also being full fat MMOs.
Besides, this thread is about VRChat, which is definitely not an action game.
These is easy when a couple seconds latency is acceptable. In an action game, as opposed to an MMO, it isn't. 100ms is likely to lead to a very subpar experience.
Thus you end up doing stuff client side, even if you sync up with the server a couple times a second.