You could just do the same thing that the Torque pack did, basically.
Have one person be the server - they are responsible for actually saving the level.
Let clients connect, and replicate their camera position and such, as well as any objects they create. It would also need some capability to upload resources to the server when they’re imported, but that’s not too crazy.
Clients make objects, their info is replicated to the server. The server saves the map with any objects created on the local client and any remote clients.
No need for an intermediary server to host everything, really. Just need to be able to replicate the actions of clients in the editor to a server.
Obviously some stuff would be disabled as a remote client - baking the lighting, playing, etc.