That's only an scalar code optimization though. You can still do comparison in constant time on little endian data, you just lack the ability to early-exit from the loop. In the modern world of SIMD data, that optimization often doesn't exist. Fair enough point though.