Still depends. If the indices are pointing to a dense, flat, contiguous array, it will still be faster than following pointers into scattered heap allocations, with or without prefetching, because of how CPU caching works.