Here’s an alternative hypothesis: it works well in the lab because the problems are small and well-defined, so that while the subjects’ performance seems to be a result of TDD, it is actually because they have a fairly clear mental picture of the entire problem and their intended solutions, so the first pass does not need a whole lot of refactoring, and the changes are incremental. Without that guiding vision, development becomes blind trial-and-error, and inconsistencies develop between the various components – inconsistencies that do not emerge until late in the game. Refactoring becomes a nightmare where each change has unintended consequences, and the roots of the problems extend deep into the architecture, and that is why refactoring can’t save the situation.
Whether you want dry-aged meat or not is a matter of personal choice. I personally love the funky, blue cheese notes of a very old dry-aged steak and am willing to shell out the extra 20 to 25% it costs. Others prefer the cleaner flavor of fresh beef. In blind taste tests I've performed, most folks prefer the slightly funky flavor and tenderness of steak aged between 28 to 45 days (any less than that and there is no discernible flavor difference). Some folks like to go even further. At 60 days, you can expect some major blue cheese funk, and it goes up from there.