I was confused about this as well, but I realized it the other day.
Mipmaps are enabled, but mipmap streaming is not.
So currently, it will use mipmaps (which is mainly good for making textures look good and not choppy/pixelated/flickering at a distance), but they will all be loaded in VRAM.
Mipmap streaming allows the higher quality versions to be unloaded if that mipmap level is not required and VRAM is needed. So in the case that there are a lot of avatars in a single area, instead of dropping to 3-6 FPS like I usually do, it would just force the lower mipmap levels to compensate at the cost of less visual quality.