P.S. It is also possible that there is something inaccurate about the way I am calculating the median article size. What I do is take the article count from the front page (say 38000), divide by 2 (19000), and then go to the special page "long articles" and look up the size of the 19000th longest article.

"Long articles" lists every page in the main namespace; the count on the front page only counts "comma" articles. Since there are probably a few non-comma articles that rank above 38000th in length, this method may actually give you a number slightly higher than the real median size of "comma" articles.