Arrays do not support all common operations on collections (and hashes don't support all common operations on maps, either). That's why e.g. C++, (which does have support for Perl-like arrays as part of the standard language, VSarkiss; vector<T> is defined in the standard!), also supports other data structures.

One thing array can't support is constant-time insertion (and extraction) of one or more elements. With a linked list, it's easy, but Perl's splice operation is quite expensive. Think of splice on a linked list: it works in constant time!

Same thing for hashes and balanced trees. C++ has the latter (as map<T>); the lack of the former is a known deficiency (most STL implementations support some hash_map<T>, but it is not standard, so precise semantics may change). Why do you need both data structures?

Easy. Hashes give you excellent expected time behaviour. But their worst case behaviour can be appalling (unless you store a balanced tree at each big bucket, in which case you're back to climbing trees). Hashes also don't support range operations ("return all values with keys between ariels and aristotle"). Some times you need the one, other times the other. This is why we have more than one type of data structure.

Perl provides some data structures. This helps make it a Swiss army chainsaw. But don't let's kid ourselves we don't need what Perl doesn't have...