Phil blogged about DebConf12 having many fewer attendees than past DebConfs (10 and 11). Dealing with registration for several years, I can say these numbers tell a different story.
First and foremost, these numbers tell me that DebConf12 hasn't happened yet, DC10 and DC11 have. People are told to sign up register early and reconfirm, but in reality, only the people who have accommodations with us have a significant motivation to do it. Once DebConf starts, lots of day-attendees will drop by, be told to register, and greatly inflate our numbers without staying for a very long time. The number of advance registrants also depends on how much advertisement is done. In terms of people who have both dates in the registration system (Phil's updated graph), I don't place much weight on this number, either.
It's my opinion that number of beds occupied is a better indication of the scale of a DebConf, and especially at this stage. This also roughly correlates to number of Debian developers staying, though for years with cheap rooms or lots of external funding, even this number can get inflated. We need some things like t-shirts and badges for every attendee (possibly registered before a certain time), but for now, the number of rooms is what matters.
In fact, I wouldn't be surprised if DebConf13 had lower "reconfirmed" numbers, because it is further away form population centers, yet still was a "bigger" conference because there are more people staying on-site or coming for multiple days.
For DebConf12, I see an average of 130 people/night, for DebConf11, I see an average of 240 people/night, and for DebConf10, I see an average of 200 people/night. So, DebConf12 is smaller, but in a different way than it seems, and even this is subject to change or variable interpretation. I'm not really wanting to predict DebConf sizes at this stage.
So basically, one does not simply expect DebConf registration data to make sense. One should not simply dismiss any DebConf before it happens. And most importantly, one can not simply use this data to plan a conference. The registration team has been frantically running around trying to correct every possible inconsistency you can imagine.