As I note on bug 34014 and bug 34549, if you've got the wrong encoding on your Z39.50 server, you can wind up importing MARC data that contains invalid data (like control characters). During copy cataloguing, there should be a warning that there's a problem with the data (possibly because of Z39.50 server config). That way we have librarians raising the flag that there's a problem before the data is in Koha rather than us doing cleanup after it's been imported. (Of course, bulk loaded MARC will still be a problem, but baby steps...)