By your logic we should have stayed with 16-bit. 64 bits enabled optimizations in calculations that have nothing to do with using numbers > 4x10^9, it has to do with having a single instruction that can operate on 64-bits (or 2 32-bit integers), etc.