If A sends mail to B and something goes wrong it is A who should be
notified, not B;
It is perfectly pointless to try to send the message in a mutilated form to B.
It is A who should fix the problem and then try to send the message again.
Furthermore, infected messages usually do not carry any sensible information and are sent without the 'senders' knowledge. Distributing an empty shell which once contained a virus is therefore just as annoying as spam.
Error messages should always contain the complete headers of the offending message. Without this, one can't tell whether the 'error message' is the result of;
Decisions about wether an email message is accepted or not should be made
BEFORE the end of the DATA phase;
If you accept a message, then change your mind and bounce, you are harassing
and should be blacklisted.