The issue is that the UO Random Number Generator is VERY prone to streaks of similar numbers in the short term - at ANY point in the curve. Draconi figured it out after we finally explained it to him with observational data that allowed him to actually test for it. Before then, they just ran tests of 100k-10 million samples, and looked at the total results, not local tendencies within the sample. And, with the streaking potentially happening at all points, that meant that a sample of 100k or more "pulls" was going to have streaks at all levels, making the streaks disappear into the mean (as every result streaked at some point).
Draconi was working on a fix on his own time when he was laid off in 2009, and we've not been able to convince the Devs after him that the problem exists, as they just run the big tests and not look at smaller subsets within it.
However, one of the ways fishermen have learned to "game" the broken system is to get a bunch of MIBs (about 50, in a pouch or crate in your pack) ,then open about 10-20. If you don't get an Ancient, stop there. If you get an Ancient, or add 5 more MIB to open each time you get an ancient until you go 10 without one, then stop). Most times, when I've done this, I'll get an ASOS, then there's about a 50/50 chance of either no AOS in the next 10, or 2-4 more in the next 20-25. (the latter which should be extremely rare if there wasn't streaking going on).