>In addition - after the above, since this is rrunnig as a sequential process, my guess is that you may have another performanc increase, if you ran the reading and the writing asynchronously, so that you can start the next read of 16k while writing the previous 16k

That is also another interesting approach. Thanks for sharing it.