8 ms·

llvm-libc's memcpy is heavily sized optimized for this reason. A dedicated instruction that is fast for all cases, is ideal, though.