Question about origin serialization

Section 5.3 has algorithms for serializing origins which include:

   "If the origin in question is not a scheme/host/port tuple, then
    return the empty string and abort these steps."

This seems suboptimal to me, since it loses information.  Better would 
be to create a serialization of the globally unique identifier if that's 
what the origin is.  The spec need not mandate how this is done, as long 
as the serialization has a good chance of really being globally unique.

Is there a strong reason for the current behavior?

-Boris