Your frequent, insightful critiques of the weaknesses in mainstream Ph.D. curriculum's has made me ponder this (you do have this question coming): if you were to put together a Ph.D. program, what would be (1) essential courses--topics and skills--that you would emphasize in the core, (2) areas of specialization and research paradigms you would have in it. What would you consider a good dissertation?
Taking as given the way that publishing, hiring and tenure decisions are made, I do not think that there is much leeway to alter the curriculum. If the basic unit of measurement is the journal article, and what the journals value is mathematical models and econometrics, then it makes sense to focus teaching on mathematical models and econometrics.
Before you change the curriculum, you have to change the incentive structure. For empirical research, we need to figure out how to measure and evaluate reliability of results--statistical significance is not an adequate metric. The standards for rigor in theoretical work have to focus on clarity of writing, rather than the use of mathematical notation. Maybe 50 years ago there was a strong positive correlation between clarity and formalism, but we've reached the point where that correlation is reversed.
Also, the weighting scheme among books, journal articles, and "other" (e.g., blogs) needs to change. Right now, a book counts for no more than a journal article, and perhaps for less. I think that a blog that demonstrates an ability to articulate economic concepts might even deserve a positive weight at some point.
With a different incentive structure in place in academia, one would be free to reform the curriculum. I would move some of the more mathematical parts of the required micro theory sequence into a separate elective sequence called "mathematical economics." I would eliminate the required macro theory sequence. I would instead require "history of economic thought," and I would put a lot of macro into that sequence. I would certainly let someone get a degree without having to master the mathematical grinding that masquerades as contemporary macro theory.
I think that a course in "communication of economic ideas" could be useful, provided that it were rigorous. The goal would be to have students be better able to think, explain, and write more clearly. By rigorous, I mean that it should have some philosophy of science perspective, so that students stop and think about when math helps to discipline thinking and when it does not. It should have some cognitive psychology perspective, so that economists can learn how they might better communicate to broader audiences. If such a course were of high quality, I would make it required.
I would be inclined to require a course in economic history. There is much to be learned from trying to understand events like the Industrial Revolution or the Great Depression. There is much to be learned from more detailed analysis of lesser issues, as well.
For most courses, if not all, I would require fewer exams and problem sets. Instead, I would require more papers, and perhaps blogs. The papers can be critical summaries of part of the existing literature, for example.
For econometrics, I think that the focus should be on the proper handling of data. Often, we take data from other sources, and the construction of that data affects how it can be used. Sophisticated estimation techniques should be taught if and only if they address unavoidable quirks in data construction. In this respect, however, I think that economists do rather well in comparison with other disciplines.
Speaking of other disciplines, it might not be a bad idea to have a required course or two in "other social sciences." Relevant psychology, sociology, and political science, all at a graduate level of sophistication, would be included.