It might be nice if you could do this, since index and substr should be faster than regexen:

$text = join("\n",@material);

for my $phrase (@key_phrases) {
  my $pos = index $text, $phrase;
  if ($pos == -1) { warn "Phrase: $phrase not found\n"; }
  substr($text, $pos, 0) = '<B>';
  substr($text, ($pos+length($phrase)), 0) = '</B>';
}
[download]

However you quite likely need to ignore case, and judging by your code you have other conditions which this code won't do. Of course you might be able to fix that with use of lc and maybe some preprocessing, but if your doing that you might destroy any speed gains.