Both methods are obviously going to work, but the difference in them is mostly about maintainability.
Using method 2 might allow you to save some in bandwidth (less repeating information) but I truly doubt the Amazon costs savings can offset the effort / man hours spend in implementing/extending/maintaining this structure.
Plus, if you ever expand this system, you have the real danger of actually hitting that 1024 limit, which (though improbably) is a risk you shouldn't take for marginal savings.