You’re right; it’s not necessarily a good idea to implement all your data structures with the same three broadly-defined types. What interests me about Perl is that you can represent almost any structure as a combination of those simple types. Tied data structures are a great example, and they make some things really easy.

If all you have is a hammer, everything looks like a nail.
I agree that the data types you mention are very useful and convenient, but I think building all data structures only with them is not a good idea. For more complex requirements, it often results in unnecessarily complicated, nested data structures. If do you go that route, though, there’s no need for arrays: there are several languages doing fine with only hashes and scalars.

Using generic data structures for everything also makes it harder to enforce particular constraints (e.g. on the type of data they should hold). Judicious use of specialized data structures (by creating new structs/classes) also makes code clearer to read and separates the interface from the underlying implementation. Incidentally, Perl also allows by using tied data structures (hashes/arrays), though that feature isn’t used very often (due to the way it is implemented).

http://people.csail.mit.edu/jrb/goo/

http://www-cgi.cs.cmu.edu/afs/cs/project/ai-repository/ai/lang/lisp/code/ext/tables/0.html