The (few, simple) WebGL examples I've looked at seem to use JSON to transport the assets from the server to the browser, but the particular representation used is apparently arbitrary.
I admit I'm thinking by analogy, I don't see OpenGL itself having helped standardise asset formats, so I'm wondering if WebGL is different here.
glTF is basd on Collada pretty closely I understand.
Generally runtime formats are designed to optimize out all the cruft that interchange formats like COLLADA and FBX have to keep around in order to be portable.
This is the reason most engines (including ours, PlayCanvas) use custom model formats at runtime. They are specifically designed to work well in a particular engine.
In Autodesk's defence the FBX SDK loads COLLADA and OBJ files very well, so we're able to support all these formats easily. The downside is that other software (such as Blender) has poor FBX support because they can't (or don't want to) use Autodesk's SDK.
I have to say that I don't seen Pixar's USD being useful to PlayCanvas as PlayCanvas is focused on real-time, where as USD is focused on huge complex scene transfers. Maybe you could take a USD and bake it to static assets but that would be about it. I say this because our Alembic tool set (which is related to USD) is barely used in the games industry (and never used as a run-time format), although it is incredibly popular in the films/animation industry.