[I guess we don't have tech-cm, so staying here.]

  Looking at "binutils" more closely, the ".gmo" files that caught your 
  attention *are* *not* from the latest import, but an older one, so 
  that seems to lend support for your idea of trimming the ".cvsignore" 
  files with a script before importing.

OK, and perhaps they should be 'cvs rm'd from the NetBSD repository.

  Another data point -- the latest "unstable" version of "cvs" has an 
  "-X" option to the "import" command, the effect of which is to always 
  give the developer the suggestion to merge with "-j... -j..."

I hadn't realized that this hint was only given if there are
conflicts.  My view is that correct importing procedure is to do the
merge checkout and commit regardless (and hence catch the R files).  I
noticed that cvs now suggests using the point tag from the import as
the second -j and the previous one as the first, instead of
-jbranch:yesterday -jbranch, and I've long used explicit tags.  I'm
not clear on where NetBSD's procedures for importing third-party
sources are documented, but perhaps doing the merge step (with
explicit point tags) regardless of conflicts should be added.

It's arguably a bug that cvs import doesn't do the removal of
no-longer-present files itself as part of the import.