It's simply impossible to isolate and control all of the important variables involved in developing any software of any significant size. There's a reason for the lack of productivity data for programming in general.
Do you think if it were repeated today with a recent version of the JDK Java development time might be less? Probably not as low as Lisp but things have improved in Java land since 1.2.
That synopsis of a study is nice but I'm still not convinced all the important variables were isolated.
There are a dozen reasons the data would be bogus but I only need to state the most obvious: you can't use the same person in 2 comparable tests because the tasks would be either too trivial or too similar; likewise you cannot use different people in groups for the tests because people are too different.
Smart and dumb people have succeeded with all kinds of languages on all kinds of platforms that are widely viewed as amazing and terrible and everywhere in between, on all kinds of projects.
Surely, there has to be at least some evidence in it's 50 years? Especially given the claimed huge effect size.