In general, statistical accuracy increases with the square root of sample size. Doubling your sensitivity and accuracy quadruples your cost. It's a power law, not a normal distribution, and it pushes us toward extremes.
An excellent interview of Allan Savory by Jonathan Teller-Elsberg deals with the difference between reductionist research and process-oriented management, and brittle and nonbrittle environments. Savory discusses why reductionist research and conventional paradigms limit innovation.
Filmmaker John Liu has documented the World Bank's $500 million loess plateau watershed rehabilitation project since it began in 1995. He has made at least two compelling films about the project, including a 22-minute version was shown at the recent Copenhagen climate conference, and a more detailed 52-minute version.