Probably all editorial from the get-go, but change to final bullet some might feel is non-editorial so I did not mark it editorial.

Original text:
The URI Scheme is based on the anyURI type defined in the XML schema specification [XML Schema Datatypes]. References that are represented using the URI scheme MUST be implemented by using the sml:uri global element as a child of reference elements, i.e., elements for which sml:ref="true".  More precisely, if a model author chooses to represent references using the URI scheme,
    *      It MUST represent each reference using an instance of the sml:uri global element declaration as a child of the reference element.  
    *      It MUST treat each instance of  the sml:uri global element declaration, whose parent element is a reference element, as a reference represented in the URI scheme.

Problems:
(1) There is a danger that some readers will see
"More precisely, if a model author chooses to represent references using the URI scheme,
    *     It MUST represent each reference using..."
and conclude that (in an example document containing 9 sml refs) the choice is between all 9 using the URI scheme and 0 using it.

(2) This text also mixes concerns of the "model author" (what one might call a producer) with those of a validator (more generally, consumer).

(3) This text re-states the definition of an SML reference, and then does so incompletely.  This could be read to mean that sml:ref="1" should not be recognized as an instance of the URI scheme (which I think is NOT our intent).

Proposal to clarify:
(1) no change:
The URI Scheme is based on the anyURI type defined in the XML schema specification [XML Schema Datatypes]. 

(2) from:
References that are represented using the URI scheme MUST be implemented by using the sml:uri global element as a child of reference elements, i.e., elements for which sml:ref="true".  

to: i.e., remove final clause and insert SML before ref, plural to singular

Each reference represented using the URI scheme MUST be implemented by using the sml:uri global element as a child of an SML reference element.  

(3)
from "if                     the model author chooses to represent 
      references using the URI scheme,..."
to   "for each SML reference the model author chooses to represent 
                 using the URI scheme,..."

(4)
from "* It MUST represent each reference  ..."
to   "* The reference MUST be represented ..."

(5) Get rid of the bulleting to separate the concerns of producing a model document containing instances of the URI scheme from consuming such a document

(6) Rename "URI scheme" to "SML URI Scheme" to be crystal clear.  It is not difficult to imagine the possibility that other URI ref schemes might be spec'd later, this is just pre-emptive disambiguation.

(7) Reword the old second bullet, getting rid of It (which de-ref'd back to "model author" and would be redundant w/ first bullet)
from
* It MUST treat each instance of  the sml:uri global element declaration, whose parent element is a reference element, as a reference represented in the URI scheme.
to
SML processors that recognize the SML URI Scheme MUST treat each instance of  the sml:uri global element declaration whose parent element is a reference element as a reference represented in the URI scheme.

New consolidated text:
The SML URI Scheme is based on the anyURI type defined in the XML schema specification [XML Schema Datatypes]. Each reference represented using the URI scheme MUST be implemented by using the sml:uri global element as a child of an SML reference element.  More precisely, for each SML reference the model author chooses to represent using the URI scheme, the reference MUST be represented using an instance of the sml:uri global element declaration as a child of the reference element.  

SML processors that recognize the SML URI Scheme MUST treat each instance of  the sml:uri global element declaration whose parent element is a reference element as a reference represented in the URI scheme.