The thing is, each document must be read from the Kubo store on the disk, which only yields 11 documents per second on my machines. As we parallelize the process by querying multiple peers ...