In my mind, there are two main differences between those chunks of code (ignoring the differences in functionality). First, the algorithm is different. Second, the top one chains expressions. I'll rewrite the top one without the chaining:

sub YetAnotherMakePtag {
  my $fixme = shift;
  my @lines = split "\r\n", $fixme;
  my @nonblank_lines = grep { /\S/ } @lines;
  my @wrapped_lines = map { "<p>$_</p>" } @nonblank_lines;
  return join("\r\n", @wrapped_lines;
}
[download]

I would argue that this version is clearer than either original, with or without comments. (Probably the slowest, too.) While the chaining made things more confusing, the algorithm change actually made things clearer. And that's because the concepts in the new version, however many there may be, are a direct mapping to a natural conception of how to perform the task. In MakePtag(), there is an intuitive leap required to see that you can wrap delimiters around a string by replacing separators with close-delimiter open-delimiter pairs, and then wrapping the whole thing with open-close.

I try to avoid that construct in my code. For example, I always do print "$_\n" foreach (@list) instead of print join("\n", @list), "\n" simply because the first is a more natural translation the actions I wish to perform. (It also reduces repetition of the terminator, and is much more open to modification if you want to do something a little different: print "-->$_<--\n" foreach...).

On the other hand, for pedalogical purposes I think Ovid's is best. It answers your question, it gives a more complete solution, and it introduces a useful technique (chaining) that doesn't particularly obscure the clarity of the example because it's so easy to see the transformation I did above.