As Robert and Ken pointed out, one explanation could be that the
content is converted twice, the second time incorrectly.
I saw those replies, but I wasn't sure how to interpret them (as in, the
evidence is compelling, but I have no idea why that would be happening or
what to do about it).
That was me just suggesting a possible mechanism to explain how we got
from a single ISO-8859-1 character to six bytes.
I am wondering ... do you maybe have some old configuration in mhn.defaults
or your .mh_profile that does some iso8859-1 to UTF-8 conversion?
--Ken