lhoward wrote Shift, Pop, Unshift and Push with Impunity!, which gives you a shallow overview. There also is Perlguts Illustrated, which tells you about the Perl data types and how they are implemented at the C level.

Whenever it comes to optimizing Perl, the most time is gained by making the program do less work. This means, use an optimal algorithm instead of trying to micro-optimize how you use arrays or hashes.

If you want to look in the direction of micro-optimizations, first do a benchmark with Devel::Dprof before trying to guess where your program spends most of its time.