From my experience with their competitor, VSA, I'd say that experience and hands-on practice are going to be the best training. These tools are more like FEA, where the user needs to understand the underlying methods and is capable of doing the task without the software, just more slowly. After that, it's just working with the program until some level of speed is reached.
I have mixed feelings about the Monte Carlo complaints. While the purely algebraic method offers some assurance, the real product will be produced in a real factory and will be more in line with the Monte Carlo method. In addition, at least under the VSA system, the Monte Carlo approach was extendible to external software, allowing analysis of non-geometric effects.
For me, the largest problem is that to make it worthwhile actual factory production capability needs to be known. It doesn't matter so much what tolerances are allowed for on drawings as what variations will be produced.