To understand what I mean with a compromise between 1 and 2, consider
this situation:
We require RFCs to be modifiable, but invent a new document serie "RFC
Attachments" which is permitted to contain non-free material, because
such material may be useful for implementers. The attachments could
be stored in a file RFC4711-foobar.txt. Then, it would be legal to
include the RFC itself in free software, as long as you don't also
store the RFC attachment in the free software.
This approach would solve the problem Harald mentioned that, to the
extent this really is a problem, it is difficult to read and implement
an RFC without exposing yourself to reading copyrighted code that you
can't use.
Of course, this was a strawman that may have obvious problems, but I
don't want to assume that it is impossible to solve those problems.
/Simon
[1] I consider the assumption that we must distinguish between code
and text a mistake, people have attempted to derive useful definitions
of code and text and failed before. Until the WG is approaching a
solution to this problem, I'd continue advocating a simpler solution
(entire modifiable RFCs) because I believe we won't find a good
solution to the problem of separating code and text.