It is quite easy to solve the problem permanently. Just use another fast structure, like a BTREE, and the possibility of pathological cases becomes impossible. Alternately one can use a hash, and then add logic to detect when a hash bucket has an abnormally long linked list, then switch it on the fly to a more robust structure.
in reply to hash collision DOS
The problem is that the average performance of hashes are very, very hard to beat, and any kind of paranoid safety logic will be hit all of the time, slowing Perl down a bit on virtually every step. Odds are that nobody in the wild has actually been hurt by Perl's hashing behaviour (yet). I guarantee that people will wonder why Perl slowed down overall. When do you make the engineering decision that average case performance matters more this time than worst case performance?