Performing science as a part of the global scientific community – meaning choosing your own research agenda, publishing your results openly, and receiving prestige based on primacy of those results – is intuitively valuable to intelligent scientists. Who wouldn’t rather be a research scientist instead of some unheard-of engineer? But knowing how valuable it is to have such freedom is difficult to calculate. On the one hand, firms can pay scientists less if they have a preference for science. On the other hand, if access to the scientific community is valuable to firms, and if potential scientists can partially expropriate that value from their firm, then simply measuring pay for research vs non-research scientists will not capture pure preference for the science job among the workers. Beyond this issue are the standard selection problems: maybe research scientists are more capable than non-research scientists, or conversely, maybe non-research scientists take lower paying jobs because the potential for career advancement is better than that of research scientists.
Scott Stern attempts to get around these issues by surveying PhD biologists on the job market. The biology job market generally results in multiple job offers per candidate, and many candidates have competing offers from research and non-research jobs, both in and out of academia. Controlling for individual fixed effects – this is where we need the multiple offers per candidate – Stern finds the average worker is taking a 20% pay cut to work in a research job (one which permits publication of results). This result is roughly robust to removing academic jobs from the sample, to considering only accepted jobs and not just offers, and to an alternative specification where workers are interviewed and asked to rank their job offers from best to worse among a number of categories (ability to follow current research program, monetary rewards, etc.).
My intuition (perhaps incorrect as I do not know of any empirical work on this topic) is that over time, more and more people are moving into jobs where the job itself provides utility at some level, rather than strict disutility; call this “reverse alienation”, perhaps? This actually has massive relevance for policy. For instance, as Stern notes, to the extent that scientists are willing to produce innovation at lower wages when “the scientific community” or “publication priority” is the reward, then strengthening IP or market incentives for innovation may be socially harmful. I think such effects are actually widespread in the labor market, and surely they should be no surprise to academic economists, nearly all of whom are taking huge discounts on what their non-research, or non-academic, salary would be. There are limits to this idea, of course. The 20th century was pretty clear in that market incentives are often far stronger than culturally-induced preferences; just look at crop output differences in China between the Deng era and the Cultural Revolution. Nonetheless…
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.159.6981&rep=rep1&type=pdf (NBER WP – final version published in Management Science)
A Fine Theorem 
I think there’s a fairly clear parallel with software development, where I can think of plenty of people who could – right now, due to particular market circumstances – command large consultancy fees but prefer to work for less money on projects that are personally interesting to them. These aren’t rich dilletantes, although they’re probably above-average in socioeconomic terms (probably a similar level to the scientists). Many of these projects are open source and are thus “given away”, further reinforcing the notion that economic incentives aren’t trumping all other considerations here.