Preface: What I am about to say is a complete and utter assumption.

One would assume that ATT did some analysis on bandwidth usage patterns amongst heavy users. They probably came to the conclusion that the top 5% of bandwidth hogs account for X% of their entire bandwidth throughput.

ATT's approach doesnt really solve their problem very much though. If someone has extremely sporadic bandwidth usage but uses gigs at a time, limiting these users won't do much for solving whatever problem they are trying to solve. Maybe their data modeling/forecasting is so accurate that they are able to segment those 5% into various segments and limit that way.