Hello,
If I'm not mistaken, the random page scheme makes some pages much more likely to be retrieved than others. If I understand correctly, each page is assigned a random number (cur_random) when it's created. To retrieve a random page, a random number x is generated and the page with cur_random > x and lower than any other cur_random is returned.
The probability of retrieving a page is proportional to the gap between its cur_random and the next lower cur_random. If cur_random is generated uniformly, the gaps have an exponential distribution. This implies there will be some gaps that are much larger than others. I guess that we could directly test this by dumping out cur_random and computing the gaps.
Since the bias (if it exists) is independent of everything else (category, age, size, authors, etc etc) I think it would be no more than a curiosity, except for people who are doing some kind of comprehensive statistical stuff.
Sorry if this is just a rerun of some earlier discussion. I did look but couldn't find anything.
best, Robert Dodier