How do NoSQL databases allow you to do a better job at "secondary data reuse" than a traditional RDBMS? Is it just a matter of (1) Denormalization (2) performance gains from not needing to worry about transactions?
The flexibility of the NoSQL model coupled with the exposure of your data to the m/r paradigm is the big win from my vantage. Almost every NoSQL solution will expose your data via bindings in virtually every programming language allowing almost any programmer to leverage NoSQL. Now you kinda have to have experience with data warehousing. Cost wise, what you would spend in licensing can be invested in hardware but more specifically talent instead.
Secondary data reuse as a concept is highly unstructured. Sure your primary systems capture data in specific formats but your analysis can take you in all kinds of directions. Being able to slice and dice without having to pre-define will be huge.