I just realized that RSET resets $rcpt_fail_count, thus making any
delays after a certain number of failed recipients useless. Is there any
objection against changing that or at least introducing a new variable
with the total number of failed recipients per session?