John,
 
This will work then based on knowing that PI's can be scoped. I don't even need to build a PI table to parse this design since a "Current PI" WILL ALWAYS MATCH the element to follow.   This avoid XML's un-orderable attributes, um problem.  
 
It adds more markup in terms of data size.  It takes away (a little) from the "beautiful" readability of XML.  It WILL WORK and I would need a day to tweak XMLFoundation to support that.  One (small) drawback is that "cut and paste" of XML in an editor could be made invalid by not pairing Element PI's properly.  That cant happen with the smaller foot printed OID as #1 attribute.
 
Perhaps the #1 placed OID should have a PI equivalent to support any existing environment that cannot easily add an ordered attribute 1st prior to writing the rest of the attributes.
 
This is some good thought.  huh?
 
Brian