For such a small collection of data my hunch is that it won't matter either way. I recommend trying out implementing both patterns and running some micro-benchmarks though to see for yourself - doing that is always a useful exercise for this kind of performance question.