>There are lots of questions where the experiments
>needed to find answers would be too expensive or are impractical for
  If one language is better than another by x%, then presumably we
could begin saving x% of the total software budget, each year for
the foreseeable future, by switching to that language.  Just 1% of
just $10**9/yr * 10 years = $100 million dollars.  Surely, that
would fund even an expensive set of experiments or data gathering.
And, if x > 1%, or we're talking all software, not just DOD software,
that benefits from the study, and we're moving faster on a path to
future, bigger, improvements, then we're "starting to talk Real money".