One of the most common ways to learn a new concept is by reading an example implementation. I expect that most programmers, given a choice, prefer to see the example in as high-level a language as possible, to minimize the amount of irrelevant detail they have to ignore. I do, anyway... don't I?
Recently I saw an unfamiliar statistics algorithm illustrated in C++ — in particular, in a very basic subset of C++, with one class and no pointers or templates. To my surprise, I found that reassuring. The use of an unexpressive language made it obvious that there was nothing fancy going on — no fancy control flow, no infinite data structures, no high-order functions, no polymorphism. C++ has many potentially confusing features (e.g. templates, overloading), but in a one-page example, it was easy to see that none of them were used. It was also obvious that the algorithm didn't depend on any exotic language features like lambda or laziness or bignums or garbage collection, so I could easily use it in any language.
A more powerful language might have expressed the example in less code, but not a lot less — it was largely arithmetic, which C++ is not bad at. And in a more powerful language, even a smaller amount of code might have expressed many more plausible things. The choice of an obviously unexpressive language ruled out large classes of things the example wasn't doing, allowing me to more quickly understand it without getting lost in a large space of possibilities.
On the other hand, I have already forgotten the algorithm, so maybe I didn't understand it very well after all.