Jeff, this feels very much like a straw person argument. None of the examples that have been held up (Google's knowledge cards, Bibliothèque nationale de France, our simple cards) link everything to everything.
I do appreciate this discussion, though, so let me pose something to you as a cataloger. Might there be a new role for cataloging here? Specifically, I see a possibility in a Linked Data environment in which the cataloger takes on the metadata
equivalent to the role of the bibliographer or selector. Rather that managing metadata through the direct editing of catalog records (irrespective of supervision/delegation), a cataloger curates a series of algorithms that define how information is retrieved
and used locally at the point of need. A selector does not read every title purchased despite some patrons' thoughts to the contrary.
For example, it could be the policy of a given cataloging department to trust or not trust data points from Wikipedia derived Linked Data sets. Or it needn't be so stark: to trust only
certain data elements from DBpedia or Wikidata (just the facts, please!). Or maybe given an Art History related call number range for a title, check to see if Getty Vocabularies has any info about the main or added entries? Or given a sound
recording material type, is this an album and does that MusicBrainz has more detailed track and performer info than my MARC records?
It is not hard to imagine an interface in which these kinds of trusted data source negotiations are managed simply by using URI domains and ontology/vocabulary predicates. It is a rather profound shift admittedly to move from the model of total
control when editing/loading every record locally to something that feels like a Bayesian probabilistic model of algorithmic metadata management. But the major paradigm shift here may be in the role of staff rather than the focus on RDF triples and the way
we store the bytes.
There is a natural transition path or roadmap if we focus on authority control and Linked Data rather than just merely shifting to RDF serializations. Again, to bring the discussion back to the list topic, this would suggest that BIBFRAME will
be more successful to the extent that it is the RDF foundation for an extensible metadata environment. Pragmatically, a BIBFRAME'd catalog should include sameAs assertions, i.e., it should include actual links, but smartly chosen ones.