One of the main questions is, how will the compulsory license revenues be distributed. In Fisher's model, there will be some sort of counting done. Since I have been working on a voluntary version of this for some time in order to help me keep track of what I listen to and listen more often to tunes I don't hear as much, I can certainly attest to how difficult it is to do rigorously. So, Seth's point is certainly well taken. The devil is in the details and 100% measurement sounds Fritzian, (or Orwellian). However, on many platforms it is relatively easy -- hey if I can code it, it must be easy. Particularly if you don't have to worry about cheating. So, if we aren't worried about being accurate, but only about sampling and estimating, I think it can be pretty easily done (see Neilson's).
Update: Ed Felten has picked up the thread with a similar but different post which takes into account why people might want to cheat.
See also my comments on Seth's reply.