Firstly what are the 3 points? Simply the best case, the worst case and the most likely. Just like when you ask how long the repair work will last and you get the reply - "I think it'll take 5 days but if we're lucky it could be 4. Or maybe it'll be much more difficult than I think. It might even take 10 days if there are unexpected difficulties."
In this case the person carrying out the work has given you more information than simply saying it'll probably take 5 days. We have a handle on the anticipated risk associated with this task. The risk associated with the estimate is captured in the gap between best and worst cases. A very wide range shows that he really has no idea how much effort is required for that
The nice thing about giving xProcess 3PE's is that it give you back 3 potential end dates - the 50% end date (you're as likely to finish by then as to overrun), the 75% end date (just a 25% estimated risk of overrrun) and the 95% end date (overrun risk down to just 5%). It give you as a manager or custmoner more information to allow you to handle contingencies. These dates are sensitive to on-going updates to information about resource availability, other task priorities, task dependencies and so on, and if these change the forecast dates change immediately.
The one caution to make about the three-point estimating technique is that it comes up with such authoritative estimates for what we know is a risky and uncertain business. xProcess is only manipulating the estimates that you give it and there are a string of assumptions involved in both the mechanism of making the estimates and the higher mathematics that are used under the covers. My conclusion? Well if you can find a bookie to give you odds of 20 to 1 that the project won't slip its 95% end date I'd take it every time! However the main thing is not just to look at the forecast end date (the 50% date). If the 75% and 95% dates are much later start planning for overruns or marshall your resources to reduce the risks.