I hear there may be opportunities to optimize throughput (or latency) when reading or writing objects by tweaking the chunk size. This is a placeholder to make sure I do not forget to ask the performance team about it.