What kind of arrays are you talking about? Associative arrays (mappings) will be more efficient since that is what a hash table is designed for. Accessing a hash table using a numeric index however is particularly inefficient since it's not what they were designed for, in which case variables would in all likelihood be as fast or faster. I haven't got the time/inclination to create and run benchmarks at the moment, why not run the benchmarks yourself? If you do, post the benchmark code you use aswell.