Adding to a bunch of other comments here, I've enjoyed combining Perl structures with BerkelyDB. As a multi-level nested data structure, I've found that MLDBM is a good source.

I do want to try and play devil's advicate for a little bit, though. I always (usuall sometimes) try to think a little harder of when I am using a hash to 'see if I can figure out a better way' and implement the same idea using an array instead. Here's my reasonings on that (and mind you, there have been only some rare occasions where I have really been able to better implement an algorithem with an array versus a Perl-ish hash.

I try to remember that a Perl hash is just an associative array. Whereas it is really an array of linked-lists. Therefore, no matter what, it can't be as fast as accessing an array directly. Not only does the key have to be hashed, but there inlies a chance it will have to iterate through n linked values before it finds the right key. Not to say this isn't blazingly fast, though not like the lightning you can pull out of an array index.

Like I said, there have been some seriously rare cases where I (or those amongst me) have been able to pull a smidgen of speed or efficiency out of an array implementation rather than a hash, although always something to put in the back of your mind.