The lecturer in our Numerical cognition course told us of a result that went along the following lines. A schoolteacher was trying to teach his students to do basic arithmetic, and seeing them get the calculations wrong time after time. Then one day he decided to follow them out into town, where he saw that some of his students handled arithmetic just fine when they were doing grocery shopping, or working part-time selling things. Inspired, he returned to class and reworded his assignments to be about shopping, and guess what happened? The students failed just as miserably as they had before. The cognitive context was just too dissimiliar to the environment where they'd picked up the practice.

I got the impression that this wasn't just an isolated anecdote, but had also been replicated in more controlled studies. The reference he gave is to Jean Lave's Cognition in Practice - I have a copy of the book from the university library, but haven't had the time to read further yet. I'll see if I can skim it through this evening to find the part he was talking about.