My current regex looks like $text =~ s/(<\/Mil.*>.*?)[\x90\x8F](.*?\x9D)/$1$2/ig; And this only removes the very first control character as expected.

Not even.

• It doesn't necessarily remove the first \x90 no \x8F because of the greedy /.*/.
• It can remove a \x90 or \x8F after the \x9D if there is neither \x90 nor \x8F in one of the blocks. /.*?/ is rarely the right tool.
• That would presume the newlines weren't there since /./ only matches newlines when the s modifier is in effect.

Anyway, here's a two-state parser that should do the trick:

my $out = '';

PROCESS:
for ($text) {
   pos() = 0;

   for (;;) {
      # Search for Mil element
      for (;;) {
         /\G ( [^<]+               ) /xgc && $out .= $1;
         /\G ( <\/Mil(?:,[^>]*])?> ) /xgc && do { $out .= $1; last };
         /\G ( <                   ) /xgc && $out .= $1;
         /\G   \z                    /xgc && last PROCESS;
      }

      # Search for end of Mil element,
      # removing \x8F and \x90 as we go along.
      for (;;) {
         /\G ( [^\x8F\x90\x9D]+    ) /xgc && $out .= $1;
         /\G   [\x8F\x90]+           /xgc;
         /\G   \x9D                  /xgc && last;
         /\G   \z                    /xgc && last PROCESS;
      }
   }
}
[download]