With the current "graceful downgrading" system, avatars' features are disabled based on their specific rank.
While I think this is a very smart system and a very positive step for VRChat, polygons are also taken into account in this system, which causes half the avatars in a room to transform into robots because they are over that limit. I hate that.
There are a couple of ways to solve this especific issue:
* Let the user define how many polygons they think is enough (this would be nice for identifying crashers and such);
* Perform some kind of automatic decimation (I assume this would be difficult to implement);
* Create another rank that is worse than "very poor" (I feel like this would incentivize people to make further unoptimized avatars, so I only see this as the laziest option);
* Use the not yet implemented techniques for quest users that were described on a recent devstream such as photogrammetry or hollowing their appearance based on a 2D camera (I personally wouldn't want this as there would be many visual artifacts on a less-than-excellent approach).