Conversely, a company that perceives no value in the grading tools may opt for an individualized session that ignores these features may never attempt to incorporate this powerful feature and lock their old inefficient workflow into a new generation of software of wasted potential.
If this ever happened, I would perceive it as a failure in the teacher. By using the standard data from the company, the instructor should be able to get a very good idea of what the company does. And if the instructor realizes that this company really needs to know how to use the grading tools and fails to teach it because they fail to ask for it, then that is a failure in the teacher. (Or are you saying that the company would argue with the teacher, and tell him "we don't need grading tools; you don't know what you're talking about, teach us something else"? If a company felt that way, why are they hiring the teacher??)
In general, what I see happening is that a company wants to transition to Civil-3D, so they hire an instructor to come in for a 3-5 day class. This is an "intro" course. And in a 3-5 day intro course, I can think of very few things where canned data would provide any significant benefit. On the other hand, for people just starting out with C3D, getting "old-style" data into C3D tends to be the single biggest hurdle they face initially. And unfortunately, the "official" training seems to echo James's feelings, that data cleanup detracts from "teaching Civil-3D". I don't think they can be separated, and getting data into C3D is a fundamental part of using C3D, so any intro course that ignores it is doing a dis-service to the students.
It's more up-in-the-air for continuing courses. For example, a typical thing is that a company will run for a while with C3D after the intro course. Then, when they feel comfortable with what they're doing, they schedule another 3-5 day training seminar to cover more-advanced topics. At this point, the instructor should already have a good idea of what the company does and how, because of the ongoing relationship. And presumably, the instructor has significant real-world experience. That means the instructor should have a pretty clear indication of what the next steps should be. And this time around, since the company is familiar with the basics of working with C3D, it will make more sense if the instructor needs to start talking about workflow changes, and new processes and procedures that are significantly different from the company's current practices. At this point, I can actually see some value in "canned data", because there is no need to go over the basics anymore. So it makes sense to have a canned DWG containing a corridor that is mostly-built so that the class can study complex intersections, for example.
But still, I think there is greater value from using the company's own data set. By this time, the company probably has a project they are working on with a complex intersection in it, or one they recently "gave up" on modelling completely because it was too complex. So, if the instructor works with them on getting THAT intersection working, he basically covers all the same tasks he would've covered with the canned data. But again, because it's real-world data, it provides the chance to hit real-world problems that are "glossed over" by the canned data. It is far better for the company to hit this sort of problem while the instructor is present, rather than after the instructor leaves. And again, working on the company's live data gives more feedback to the instructor, giving him a better idea of what the company currently does and how, which can provide guidance for what needs to be taught.
I think what I would find most valuable is something of a combination. This would involve an instructor who came and taught the classes on our real live data, but also brought along a collection of "canned files" that illustrate complex tasks in clean ways, such as
the one Dana posted. But again, I just see so much more value realized from using the company's live data that I think it should be the initial choice. Then, if the real-world data fails to illustrate any important points, the instructor can always pull out the canned drawing. And the instructor can leave the canned drawings with the students, as a reference material for later. This is what I would find to be the most-valuable course.
Of course, one thing I haven't explicitly mentioned is that the instructor hopefully has significant real-world industry experience, and was good at it. Unfortunately, the only requirements for becoming a "certified" trainer are six months experience, and proving that you can teach the scripted class. One of the issues we've had with the "approved training" is we've seen people we know (because one or more of us previously worked with them) go get certified and become official trainers. These are people we wouldn't hire to work for us, and they go out training others. That makes us shudder. And it's probably a more-significant element in quality of training than whether or not the instructor uses canned data.