Another approach, with some attempt to make parsing more tolerant of variations in format:

c:\@Work\Perl>perl -wMstrict -le
"use Data::Dump qw(dd);
 ;;
 my $start = qr{ > \s*       }xms;
 my $more  = qr{ \G (?<! \A) }xms;
 my $post  = qr{ \s* <?      }xms;
 ;;
 for my $s (
   '<span c=\"#12\" foo=\"%DoNotMatch%\" bozz=\"false\">%PN1%</span>',
   '<span c=\"#98\" bar=\"false\" zot=\"%NoNoNo%\"> %DIMMM% %DIMINCH% 
+</span>',
   ) {
   print qq{'$s'};
   my @matches = $s =~ m{ (?: $more | $start) % ([^%]+) % $post }xmsg;
   dd \@matches;
   }
"
'<span c="#12" foo="%DoNotMatch%" bozz="false">%PN1%</span>'
["PN1"]

'<span c="#98" bar="false" zot="%NoNoNo%"> %DIMMM% %DIMINCH% </span>'
["DIMMM", "DIMINCH"]
[download]

Please see perlre, perlretut, and perlrequick. Caveat: any "pure regex" approach to parsing XML is fragile, probably very fragile.