The version 8 file contains 847,223,402 hashes which gives 29.7 bits, that is where I got the 30 reads from. And I assumed the most naïve approach, always just seeking and reading a single hash, but admittedly this would almost certainly not read the final page over and over again. And then I just subtracted 10 taking into account that you would read the final page only once, with 10 being a rough approximation of 12. How many bytes you actually read per read could be affected by all kind of things - sector size, cluster size, page size, read ahead, ... - so 10 bit was good enough for me. But you are right, the number of reads would almost certainly be in the vicinity of 20, not 30.
