I agree with Peter Bruce that resampling methods can be valuable for teaching and are valuable in practice. However, they do raise a thorny issue for teachers. Until we see them appearing in published research, I don't think we can afford to base introductory courses on them. David Moore recently made the same argument against basing the introductory course on Bayseian methods (at a discussion at the August ASA meeting). I would be similarly opposed to that or to basing it on nonparametric methods or any other specialized approach. Surveys of what statistics methods are actually used in print and in practice are quite clear; students need to know how to display data, how to summarize the patterns (and exceptions!) they see, and how to draw (frequentist-based) inferences with t-based intervals and tests for conclusions about means, differences between (and among) means, and regression slopes. They need to see contingency tables and understand chi-square, but may not need conditional probability for this. With these skills, students could read 90% of the statistics in most medical, business, social science, and science publications. (They'll still need to learn the science, of course.) There is a chicken and egg problem here; hard to change the ways of the world if we don't train students to new ways. But I can't support making our students into the new breed of chickens -- at least until things settle down enough to know which breed they should be.
We have actually seen such a change. The AP course is quite modern compared to the way statistics was taught even a decade ago (and compared to many texts still widely used.) The emphasis on graphics, on understanding, and on data exploration was considered radical when I started teaching it in the 1970's. However, it is now quite standard in practice. (Just read the ads for the major statistics packages to see what they emphasize.) I think we need to make future changes in our teaching in much the same way.