Formality: References from std-track to BCP

[Harald Tveit Alvestrand <harald@alvestrand.no>]

In reviewing draft-rfc-editor-rfc2223bis-04.txt before Last Call, I came 
across the following paragraph:

     Some standards track document use certain capitalized words
     ("MUST", "SHOULD", etc.) to specify precise requirement-levels for
     technical points.  BCP 14 (RFC 2119) [3] defines the proper
     interpretation of these capitalized words in IETF documents.  If
     these words are used and capitalized, RFC 2119 should be cited (as
     specified in RFC 2119) and included as a normative reference.  (It
     is noted that this is a formal violation of the rules of RFC 2026,
     since RFC 2119 has BCP category.)


I believe this is inconsistent with both common sense and current practice, 
since we *routinely* let standards-track documents (at any level) refer to 
BCPs for things like registry process (see MIME for a prominent instance).

But is there something published that supports this view formally?
RFC 2026 is .... unclear.

                   Harald