In the so-called good example, both local $_ could easily have been replaced with my. You gain the readability of a named variable at the tiny cost of having to specify the topic of two operations.

my @a = map { s/\s*(#.*)?$//g; $_ } <DATA>;

my @b;
while (my $group = shift @a) {
   if ($group =~ m/\D(\d+)-(\d+)(?:\D|$)/) {
      my $range = "$1-$2";
      foreach my $i ($1..$2) {
         (my $addr = $group) =~ s/(\D)$range(\D|$)/$1$i$2/;
         push @a, $addr;
      }
   } else {
      push @b, $group;
   }
}

print join("\n", @b);
[download]

There are cases where using $_ helps, but I don't think this is one of them.

Note: The localization of $1 and $2 is unnecessary. It's already being done by the for.

Note: local $_ doesn't preserve $_ in all situations, unfortunately. for ($var) is a far better way of aliasing $_.

Thank you for your point regarding unnecessary localization of $1 and $2. Still, my point was avoiding intermediate (possibly unnecessary?) variables, i.e. $addr and $group in your code that incredibly looks very much like my original!

Ultimately, I guess it is style. One thing I love about Perl is TIMTOWTDI and I admit I tend to go for minimal keyboard presses unless extra performance is needed. Then, I would probably lean to doing it in C++.

Update: Incidentally, the named approach is probably a little faster than a localised one. However, only slightly!

Still, my point was avoiding intermediate (possibly unnecessary?) variables,

I did not add any new variables. You used two (both named $_) and so did I. My point was that naming two variables $_ is more confusing, less readable and less maintainable.