Is there something different about how DB is built?
This strikes me as being akin to saying "Gallup surveyed all people in Patrick's kitchen last night and estimates Obama's approval rating as 0%. Talking Head, this number is clearly in error. What do you think is different about Obama which makes it so difficult to poll about him? Is it that he is black?"
No, its that biased samples bork results. (I'm using the statistical, not pejorative, usage of the word "biased".) There are a million things that could be happening here: a userbase primarily outside of the cooperating ISPs coverage areas, a younger userbase who disproportionately do not install toolbars, etc, etc.
>Which other sites show such a large discrepancy between self-reported and quantcast/compete/alexa numbers?
I can expect a pretty small overlap between web users who have, say, the Alexa toolbar installed on their computers and the kinds of early adopters who would use a service like DailyBooth.