If you stretch your imagination enough (and Perl is good at letting that happen) you can solve these puzzles using Genetic algorithms.
You could assign each character in a string a task:
5 - fill 5 gallon bucket
3 - fill 3 gallon bucket
P - pour 5 into 3
p - pour 3 into 5
E - empty 5
e - empty 3

I don't think I'm missing anything. Your population would be strings of different lengths, maybe 3-10? I'm guessing here.

Then your fitness function takes the string, and just follows the instructions like a recipie. This does mean that some semi-impossible things could happen, empty an empty bucket, fill a full bucket etc.

Crossing is just taking random points in the string and swapping them.

Once you have a good GA engine it is easy to replace the fitness function and the population generation functions, and in a few cases the crossing function (mutation too!).

If I get bored sometime today maybe I'll code up an example.

Doing a search on "Genetic Algorithms" turns up a few really good nodes.
Helter