use vars qw ( $message);

$message = qq ( Foo bar <font size="7"> phrase here </font> blivits. )
+;

$message =~ s#(\d+)#($1 > 5) ? 5 : $1#eisg;

if ( $1 ) {
         print "\n After s###, \$1 is: $1 \n ...and revised \$msg is: 
+$message \n";
        }
         else 
        {
         print "\n no \$1 found \n";
}
[download]

• the $message html fragment is well-formatted... for 4.0, but display elements are now better (well, that's what the experts say) done with css
• the "sig" regex modifiers are not needed with this $message but might well be with more typical .html
• and, while matching on <font size=" before the 7> is omitted only to keep the example simple (and is easy enough to add), this is a very limited and profoundly un-extensible method -- better to use an html parser, as suggested elsewhere. For example, a web'ster using <font... may well have multiple and/or variant forms of elements in a single font tag, as, for example,
  <font color="red" size=+1...>
• and, finally, despite generous efforts by OP, I can't see how Re^2 or Re^3 can or do work ... even with "<"s replacing "["s
  ...so would welcome education.