[camccann]

If the effect size is large (i.e. the magnitude of difference between the means of lisp vs non-lisp programmers) the size of problem should be irrelevant.

Writing a "Hello, world" is going to take about the same amount of time in any language that isn't a turing tarpit. At issue is generally some sort of asymptotic advantage, not a guarantee of "language X will always be superior every time always". In fact, for any pair of one easy and one hard problem, I'm pretty confident I could find (or invent, if necessary) two languages that each beat the other for solving one of the problems.

The claimed benefits of a more expressive language generally involve things like better management of application complexity, or greater ability to modify and restructure late in the project. Hammering out a program of a scale such that the programmer can keep an accurate model of the entire thing in mind at one time, and that is only written and never modified, would be more a measure of the programmer's typing speed and the sophistication of the IDE used than anything else.