...
Statement #1 of the Linked Data Principles says "Use URIs as names for things" and statement #2 says "Use http URIs to that people can look up those names" (This includes https URIs, too!). This stresses that URIs are used for the purpose of identification as well as to initiate data transfer.
...
Why are identifiers not independent from protocol?
That currently is the way it is defined in the RDF standardThe RDF specification says that identifiers need not be protocol-based and that it is perfectly possible to use any URI as an identifier (e. g. a URN).
Linked Data, however, is about the web so we need web-actionable identifiers. Deploying something like a resolver in between breaks the web approach: you have to educate users to use the resolver instead (which btw. has other disadvantages - bad experience with resolvers with DOI because the resolvers behave so differently. BnF experience: It's not sufficient to set an ARK on everything)
...
We could assign https to new data and keep http for "old" data. However, that would be awkwardly inconsistent and hard to explain to non-LD people, particularly if the URIs are based on patterns (e. g. "Use the prefix http://example.com/data/ and add the foo-number"). And still involves http traffic.
...
Does it make sense to change them to https as long as rdf, owl, skos, xsl ... don't change? Policy of these is to wait and hope that HSTS and UIR will still become good. Another point is that the semantics of most of those vocabularies are hard-coded in the relevant tools (such as the OWL-API or Jena) so the identifiers are not dereferenced before they are acted on, e. g. for building class hierarchies using rdfs:subClassOf.
DCMI has had only one request for https so far and therefore currently doesn't see urgent need for action.
...