So many repositories have their own content models and interrelated objects. Indeed, even within one repository you may have multiple, sometimes overlapping sets of object relationships, types, or attribute/value pairs. These are recorded and sometimes enforced in a variety of ways. As a tool that fits in between repositories or data structures, the workbench needs to allow this kind of annotation, without getting bogged down in any one particular model.
While METS does make room for certain attributes and structural relationships, these are limited to xlink form and XML ID references. We have mapped such smLinks for things like surrogate/thumbnail relationships. But RDF might offer some advantages and could just as easily be embedded in METS.
So if the workbench support RDF tagging of the arranged objects, you could pick and choose the namespaces you wanted to use. Some could be custom and others could be standard, such as LOC PREMIS RDF. They can be mixed together on the same objects. Seems ideal.
This issue is really a placeholder for a compelling design idea for the software. Comment freely!