Your script is very elegant, shorter and functionally equivalent.
However, this simplicity comes at a price. The overhead of a second hash to eliminate duplicates is a heavy load on performance, especially after adding "use strict" and "-w" on top.
I compared the execution times on a 102_759 words dictionary.

$ time perl gmax_anagram.pl < words |wc
4.26user 0.07system 0:04.32elapsed 
   6427   14802  107910

$ time perl merlyn_anagram.pl < words |wc
6.35user 0.10system 0:06.45elapsed
   6427   14802  107910

(after adding use strict and -w)
$ time perl merlyn_anagram.pl < words |wc
7.44user 0.12system 0:07.55elapsed 
   6427   14802  107910
[download]

Anyway, this approach is one more thing to know, if I want to privilege a shorter script in favor of a faster one.
Thanks.

 _  _ _  _  
(_|| | |(_|><
 _|