Try modifying networked-aframe template instantiation logic and syncing position/rotation/scale/color to use InstancedMesh or BatchedMesh as an opt-in for example for avatar-template, to reduce draw calls to render all the avatars (we can have 70 avatars with uws adapter).

Probably start with InstancedMesh and simple sphere avatars using setColorAt and setMatrixAt
Also look at https://github.com/diarmidmackenzie/instanced-mesh if there is way to integrate it to support the same glb avatar with shared materials but with different colors, then from that support different glb with shared materials with BatchedMesh.