If you refer to the original article about the piracy in World of Goo, you will see the comments left by visitors alone indicate a lot of the reasons the analysis can be considered flawed. In the end, the author conceded that yes there were some assumptions that could greatly affect the result but he hoped that some of the assumptions countered each other making the estimate accurate. That simply is not a very statistically valid way to treat uncertainty. The main reason the data is flawed from my point of view is that he counts unique IPs but many ISPs or routers provide their users with many more than one IP at different times. It is likely that World of Good was pirated quite a lot, but the alarmist figure of 90% I don't accept easily. The fact it gets thrown around regularly now in debates about piracy is just annoying. As others have noted, there are so many other factors that could influence how much a game gets pirated, that the World of Goo example is just one piece of data (or should I say one datum - who uses that word anyway?) and one of questionable accuracy to be sure.
For your convenience, here is the original source:
http://2dboy.com/2008/11/13/90/
EDIT2... Here's one example of a comment that was left by a visitor:
A comment like that raises an interesting question, IMO.
e.g. Is it
more wrong to pirate a game that you intend to play a lot than one you will play for 10 minutes? If 9 players pirate and play the game for 1 hour each, where another player (maybe a fanatic

) pays for the game and plays for 9 hours, would you record that as a 90% piracy rate or a 50% piracy rate? Of course, it's obvious which number someone in the industry would choose.