If the data is always UTF-8 encoded, it might save some effort to decode that first and then look for the problem characters?

BTW, you can do all this in a single pass, if performance matters.

sub convert_to_html_entities {
   my $str= shift;
   utf8::decode($str);
   $str =~ s/[\x{201A}-\x{2122}]/ '&#'.ord($&).';' /ger;
}
[download]

You could even just wholesale replace all non-ascii characters to completely sidestep the encoding problem:

sub convert_nonascii_to_html_entities {
   my $str= shift;
   utf8::decode($str);
   $str =~ s/[^\x20-\x7E]/ '&#'.ord($&).';' /ger;
}
[download]