This is just a guess (perhaps somebody who's more experienced
with Perl compiler optimizations can help out here). Since
accessing array elements tends to be faster than accessing
hash elements, if
the Perl compiler resolves the
pseudo-hash keys to the array indexes at compile-time, it
makes building structures with named elements more efficient
at run-time than simply using hashes.
The perlref page notes that the compiler does something
to this effect for using named fields in objects. I'm not
at all familiar with objects in Perl, so I can't really
elaborate on that.