I've finished the book. Here are some parting thoughts.
First, I want to mention an important point that she made earlier in the book but I forgot to blog: Research is a classic public good, in economics jargon. Once the results of research are out there it's impossible to control who will benefit from the information, which means that the producers of the good cannot capture all of the benefit. In most cases this means that we have an incentives problem, because if people can't capture the benefits they won't produce it and then society won't benefit. Science solved this with the priority system: The first to publish something gets credit that will benefit their career and funding. This thus incentivizes people to put information out there as quickly as possible, for the benefit of themselves as well as everyone else.
Second, as far as her recommendations later in the book, I largely agree that we need a looser coupling between the production of research and the production of scientists. Ideally, all of our scientists should be produced in environments that produce research, but we don't need to produce all of our research in environments that produce scientists. To that extent, I support the idea of research institutions that don't have graduate programs. However, I would caution against using national labs as the model, as Stephan does in her final chapter. While the national labs produce great research, being federal institutions they have absurd overhead rates and procurement bureaucracies that rival Soviet central planning for inefficiency. I have no easy prescription for how to create effective research institutions that are not educational institutions but aren't federal either (a quick look at federal contracting shows that there are plenty of ways to set up bloated and inefficient systems outside of the federal government), but I do think we should be wary of encouraging DOE to metastasize even more than it already has.
Maybe one intermediate step would be to make it easier for university researchers to support technicians and staff scientists on grants. This is easier said than done, but having it as a target would keep the academic science community churning out just as much research (if not more) without the wasteful by-product of PhD over-production.
On the other hand, I am deeply skeptical of her proposal to give fellowships to students and then let graduate programs compete for them. Partly there's the fact that the programs would have an incentive to structure projects in a way that gets results out quickly and gets students out quickly (without the formative "time in the wilderness" that students spending in deep frustration but also deep learning). However, I recognize that that may reflect my biases as an old man. A bigger problem is that it would be tantamount to putting the federal funding agencies in charge of graduate student admissions. While this would certainly be great from a research standpoint (putting all sorts of information in one central place for subsequent analysis) I think there would be all sorts of political insanity in what would follow.
All in all, this book was a good, insightful read. For those who aren't academic scientists, the first several chapters do a good job of describing how the system works. For those who are academic scientists, the last few chapters (starting with the one on the job market) do a good job of raising issues that we have not grappled with as well as we could/should.