Don't know if this muddies the waters even further, but I think you actually have three coordinate systems:
the "object" (what you call "model-relative")
the "viewer", who could look "towards" the object or away from it, or sideways...
the "room" where both "object" and "viewer" are located. This is where I would use x,y,z, although I think they usually are connected with the viewer...