Actually, I think this is a bit too strong of a statement, although I intend to set up a fr-UNIMARC database soonish to see how much breakage there is.  Obviously, there are a number of tests that implicitly or explicitly assume MARC21 is in effect, and those need to be generalized, but many of the tests need only create some stub bibs and items -- and we could write a routine that replaces the various add hoc ways of doing that with a central way of getting back a stub bib record in either MARC21 or UNIMARC (or NORMARC).