So a standard reverse sort expression serves in all cases.

As I /msg'ed you at the time you posted this, I had a doubt about this - and now I've finally found the time to confirm that I wasn't being entirely paranoid ;-) Under locale, it's possible for longer strings to be sorted before shorter ones. Here are the instructions to reproduce (on a Debian-based system):

$ mkdir -v /tmp/localetest; cd /tmp/localetest
# copy "en_US" and the files it refers to via its "copy" statements
$ cp -v /usr/share/i18n/locales/{en_US,en_GB,i18n,iso14651_t1,iso14651
+_t1_common} .
$ mv -v en_US en_TESTING
[download]

Now edit the file en_TESTING and insert the following into the LC_COLLATE section after the copy "iso14651_t1" statement:

collating-element <bc> from "<U0062><U0063>"
script <TESTING>
order_start <TESTING>;forward;forward;forward;forward,position
<bc> "<a><c>";"<BAS><BAS>";"<MIN><MIN>";IGNORE
order_end
[download]

Now compile and test the locale:

$ localedef -i en_TESTING -f UTF-8 -c ./en_TESTING.UTF-8
$ LOCPATH=/tmp/localetest LC_ALL=en_TESTING.UTF-8 perl -Mlocale -le 'p
+rint for sort qw/ab a b bc/'
a
ab
bc
b
$ LOCPATH=/tmp/localetest LC_ALL=en_TESTING.UTF-8 perl -wMstrict -lMlo
+cale
print "1 $_" for "abcabbc"=~/${\join "|", reverse sort qw(b bc) }/g;
print "2 $_" for "abcabbc"=~/${\join "|", sort {length$b<=>length$a} q
+w(b bc) }/g;
__END__
1 b
1 b
1 b
2 bc
2 b
2 bc
[download]

Now of course the real locale definitions are long and I don't plan on reading all of them, and so it's entirely possible that currently no locales exist that define this kind of sorting order. But I'll just be a bit paranoid and stick with sorting by length explicitly :-)

Update: I did incorporate your suggestion about highlighting the fact that strings are matched literally, thank you!