Answer from cs61c-at (soemoe aung 16076861) for Question 3
This technique compensated for the fact that some machines lacked a guard bit for double precision floating point numbers, so there would be less round-off errors.