I can't explain why double the logarithm of the number of passwords stored would be the target. But it does make sense to have at least log_2(n) salts for n passwords. That would guarantee each password, even if they were all identical, would have a unique salt making every password hash differently (at least theoretically).
Now perhaps you would double that as a just-in-case scenario or because there is some fundamental property of every (existing or future) hash algorithm that causes a collision half the time (i.e. both salt 10001 and salt 10000 generate the same hash value).
Now perhaps you would double that as a just-in-case scenario or because there is some fundamental property of every (existing or future) hash algorithm that causes a collision half the time (i.e. both salt 10001 and salt 10000 generate the same hash value).