Named Graphs have thus far served a very useful purpose (esp. in SPARQL) for supporting notions of provenance into systems which want to store and track data from multiple sources.
The RDF next steps has sparked some discussion on the future direction of Named Graphs in the RDF standards. However, I do have some concerns about bringing Named Graphs further. To clarify, I'm highlighting concrete questions as follows:
Related question wrt. specifics of some proposed Named Graph semantics here. Do we really need to nail down a semantics for named graphs? What would this buy us in terms of current adoption/standards? Is it really useful for the prevalent provenance use-case and SPARQL? Will it enable better integration of SPARQL and RDF(S)/OWL semantics? Or is it really only necessary for multi-graph documents?
Thus far, Named Graphs have mainly been used in the context of encoding the provenance of multiple single graph documents. Besides the (non-trivial) issue of multi-graph syntax (for which reasonable non-standard solutions exist for the non-XML based syntaxes), publishing multi-graph documents for me is a bit of an unknown. It has obvious benefits for multi-graph exchange between provenance-aware agents (e.g., dumping a SPARQL endpoint, or dumping a site's content) and may be useful, e.g., for returning quads to SPARQL construct queries. However, stepping outside provenance, things get a bit murky.
Multi-graph document are an obvious (albeit not ideal) replacement for unwieldy RDF reification for speaking about (sets of) triples... useful for annotating triples with, e.g., temporal validity, probablistic information, etc.
However, multi-graph documents would constitute (as such) a new data model which is not backwards compatible with the RDF (single-graph) model—e.g., think of a graph which is annotated as expired for the current time. I understand the need for multi-graph exchange between trusting provenance-aware agents, but should we be encouraging more general publishing of such multi-graph documents? Are standard syntaxes for multi-graph documents a step too far? How can multi-graph documents co-exist with single-graph documents/tools out there now? Is there a restricted form of multi-graph document that is backwards compatible and still proves useful? Will we need to start talking about quints to track the provenance of multi-graph information?
As replacement for RDF reification, I would suggest N-Quads, where the 4th element (context) refers to the reification statement, which then mustn't include the "reification triples" (the rdfs:subject, rdfs:predicate and rdfs:object relations).
So, I would say yes, because also quads need a named graph entailement for provenance and trust ;)
PS: Further thoughs and a (probably) concrete use case can be found in "Time for quintuples?" on the Semantic Web mailing list
PPS: Documents are simply the carrier medium for graphs ;) (from my point of view). Hence, it shouldn't really care, where the graphs are placed in the cloud. It is rather more important how they are related, or? (that means, which graph refers to another one)
PPPPS: Tolle's distinction between internal context, here identified as reification statements, and external context, here identified by the Named Graph entailment and the description of the Named Graph (see also Understanding Data by their Context Using RDF)
Edit: To emphasize my still valid endorsement that quadruples are enough, please have a look at my recent proposal for optional statement identifier.