As I read it, http://bibframe.org/documentation/annotations/20130430.html attempts to solve two problems (a) lightweight representation of non-canonical bibliographic metadata and (b) knowing which RDF triples should be exchanged in inter-operability processes. I say 'As I read it' because the document appears to contain no explicit statement of the problem(s) it is attempting to address.

(a) I agree that a lightweight representation is likely to be necessary, but this is not it. Introducing semantically new RDF properties into bibframe, the current scheme introduces a complexity burden on every bibframe-compliant tool, in perpetuity. A very similar representation could have been achieved by using OWL to create syntactic sugar RDF properties (potentially in combination with selected Open Annotation, SIOC, etc. properties, if necessary). These would appear to users just like normal properties, but allow tools which didn't care about them to use only the existing properties. This would also avoid drawing an artificial line in the sand between reviews that are Works and reviews that aren't. Given that reviews range continuously from stars-of-out-three to three-volume dissertations, it's hard to imagine that every library is going to get that right every time. Syntactic sugar RDF properties would let us treat reviews as Works when we wanted to and only then.

(b) There is an already implemented and optimized technology for organizing large sets of RDF triples of divergent provenance and lifecycles, called 'Named Graphs.' As far as I can tell it is an idea solution for dealing with the fact that some of the content is 'core' bibliographic content from a consortia; some is user contributed data under a custom license; some is commercially licensed cover images; some is harvested from wikipedia; some is local cataloging of unique items; some is temporary cataloging of recently received items; etc.

cheers
stuart