Discussion started on the mailing list thread. The relevant part of the discussion (quoting from my original mail):

We generally have an issue at W3C on how we would define testing and implementation for vocabulary terms […]. The generally accepted approach is that for vocabulary items the term "implementation" is a a misnomer and it should be considered to be an alias to the term "usage" for our CR process. Indeed, there is no RS behavioral requirement for any of the SSV entries, so traditional requirements on implementations would not apply. Instead we may, for example, define "usage" of a specific vocabulary term, for our CR process, that there are at least two publishers out there who use these terms in production (hoping that at least some reading systems do something sensible with it). Such, or similar, ways for "testing" (per W3C process) for vocabularies was used in other specifications in the past.

However: would all SSV terms pass such a requirement? All the answers on John's mail suggest that the answer would be no […], because many terms have been introduced to the spec as part of an aspiration for something. Ie, we may be creating a problem for ourselves. Shouldn't we mark the SSV vocabulary as non-normative overall with, possibly, explicitly mark a few terms as normative because we know they are accepted by the community (e.g., landmarks)? Note that if we keep all the terms as normative and then we fail on the CR tests, the usual expectation would be to remove them altogether, which we do not want (I presume).