We argued over this question for quite a while and eventually put it together in this paper: http://www.ncbi.nlm.nih.gov/pubmed/11693564

The general conclusion is it depends on how long you have sampled your decay for. If you have sampled through two or three time constants (2 or three T1s or T2s) then at the longest delay the signal intensity will not have reached zero and you should fit with the two parameter equation. If you have sampled through five or more time constants then the signal intensity should be close to zero but you may have introduced an offset by the way in which you measured the intensity (by searching for a maximum in a grid about the peak center, for example) and you should then use a three parameter fit to eliminate the influence of the offset.