You need to entitize all characters not within the latin-1 reptiore. I /belive/ this code will do it, assuming $_ is your text, stored as "proper" utf8, with perl's utf-8 flag set.

 my $ord;
 s/(.)/$ord=ord $1;
 if ($ord < 128) {
  $1;
 } else {
  "&#$1;";
 }
 /ge;
[download]