The biggest failing of ID/IDREF is that the link has no way of
expressing any semantic value as to why the relationship was
established. ID/IDREFs rely on the text that is highlighted to explain
why the link was inserted. This model works if the authors understand
the semantic meaning of the links that should point to or from the file
that they were authoring. And, although this isn't important as yet,
ID/IDREF can't express bi-directional links.

I was reading up on XLinks and it appears that XLink enables you to:
(1) Link between resources without the need to change them (very imp
req in my case)
(2) Build new documents dynamically from a template link document
(could be imp as an alternative)
(3) Have true bi-directional links between resources (not imp)
(4) Group sets of related resources in strongly typed relationships
(imp)
(5) Link between structured and unstructured information (very imp)
(6) Manage large repositories of link information in a centralised
efficient manner (not imp)
(7) Link resources that are stored in a variety of different data
repositories (imp)
(8) Create links using any application that creates XML XLink documents
(not sure)

I don't know if ID/IDREF can do those or how XLink can do them, with or
without using a separate tool other than a traditional XML parser.