Friedl has a variant of this; 'commafying a number'.
I did a paraphrase of artist's solution, influenced by Friedl, but with a twist for the <br> insert. And then a cmpthese.

nacka:
  s/(\d)(?=(\d{5})+(?!\d))/${1}-/g;
  substr($_, 17, 1, '<br>');

artist:
  s/((.){5})(?!$)/$1-/g;
  s/((.){17})\-/$1\<br\>/;

thezip:
s/(\d{5}) (\d{5}) (\d{5})
  (\d{5}) (\d{5}) (\d{5})
 /$1-$2-$3<br>$4-$5-$6/x;

nacka/artist hybrid:
  s/(\d)(?=(\d{5})+(?!\d))/$1-/g;
  s/((.){17})\-/$1\<br\>/;

artist/nacka hybrid:
  s/((.){5})(?!$)/$1-/g;
  substr($_, 17, 1, '<br>');
[download]

And the Benchmark:

                   Rate    artist    thezip artist/nacka nacka/artist 
+     nacka
artist       13175231/s        --       -8%          -9%         -14% 
+      -14%
thezip       14347202/s        9%        --          -1%          -6% 
+       -7%
artist/nacka 14534884/s       10%        1%           --          -5% 
+       -6%
nacka/artist 15313936/s       16%        7%           5%           -- 
+       -0%
nacka        15384615/s       17%        7%           6%           0% 
+        --
[download]

All of the above rate numbers are probably within the error margin.

I declare thezip's solution the most readable, and hence the winning solution.