Just noticed this...
In both the original, and my version, there seems to be some wasted logic. See comments that start with ## below.

   if (length($payload)%2) {
      my $padded = substr $payload, 0, -1;
      $padded .= "\0";
      $padded .= substr $payload, -1;
      ## above: ok, so the next to the last byte 
      ## is guaranteed to be null since we just 
      ## explicitly set it
      if (ord(substr $padded, -1) >= 128) {
         my $end    = ord(substr $padded, -2, 1);
         ##
         ## above: now we're grabbing the next to the last
         ## byte, even though we already know it's a null?
         ## 
         my $endend = ord(substr $padded, -1);
         $end--;
         ## now we decrement it, knowing full well it's 
         ## going to be -1?
         $endend--;
###DEBUG: 
printf "CMPSAT = %i %i\n", $end, $endend;
         # if 0 before --, then -1 results in warning to pack below
         # -1 is in essense \xff or 255
         if ($end == -1) { $end = 255 }
         ##
         ## above,of course, it's going to == -1, and thus
         ## at this point, always reset to 255
         ##
         if ($endend == -1) { $endend = 255 }
         $padded = substr $padded, 0, -2;
         $padded .= CORE::pack('CC', $end, $endend);
      }
      $payload = $padded;
   }
[download]

It would seem then, that you could simplify all of that out, and just decrement the last byte by one?

sub prepchksum {
    my ($payload) = @_;
    printf "BEFORE = %s\n", (CORE::unpack "H*", $payload);
    if (length($payload)%2) {
      $payload =~ s#(.)$#\0$1#;
      printf "MID1   = %s\n", (CORE::unpack "H*", $payload);
      # Compensate off-by-one error
      if (ord(substr $payload, -1) >= 128) {
         $payload=~s#(.)$#pack('C',(ord($2)-1)&255)#e;
      }
   }
   printf "AFTER  = %s\n", (CORE::unpack "H*", $payload);
}
[download]

Or perhaps there's a bug somewhere, either in the 'C' implementation you based this on, or your original port?