Right now our Java solution hasn't yet figured out how to handle contractions yet. Perl is still ahead. Secondly why break out the sledgehammer when you don't have to?

foreach my $word (
  sort {
    $freq_count{$b} <=> $freq_count{$a} or $a cmp $b
  } keys %freq_count
) {
  print "$word:\t$freq{$word}\n";
}
[download]

And as for the maintainability of a Schwartzian Transform, it is a trick. If you think in a list-oriented way, or if you have ever really understood the tranform, then maintaining it when you see one is pretty easy. I understand it, and I try to ensure that people I train are able to handle list-oriented thinking. So it isn't a problem for me. But YMMV on its maintainability.

No Schwartzian Transform is required. Does Java let you easily sort ignoring case? How about by length?...

sort { $freq{$a} <=> $freq{$b} } keys %freq
sort { lc($a) cmp lc($b) || $a cmp $b } keys %freq
sort {
    length($a) <=> length($b)
 || lc($a) cmp lc($b)
 || $a cmp $b
} keys %freq
[download]

Speeding up the third case with a ST would be more difficult than using some other sort-speeding techniques (many of which have names that I don't recall). Though you'd have to have a whole lot of different words for the trade off of sort speed for code complexity to be a "win" here, especially since we are trying to write very maintainable code.