I just came across a PLoS Biology article by John Krakauer et al. in which the authors describe the transfer of motor learning across different ccontexts. I'm not done reading the article, but it looks like they trained human subjects to control a cursor on a computer screen either with their wrist or with their shoulder. When they trained them first with their shoulder, it helped them when they tested the persons with the wrist-task (transfer). However, when they trained them with the wrist and then tested them with the shoulder, no transfer was detected. Interestingly, the task was easier with the wrist than with the shoulder.
This is very similar to results I got in my PhD thesis, where the fruit fly Drosophila could only transfer color memory from the more difficult to the easier task, but not the other way around.
Some very recent preliminary results tentatively point to a common mechanism for a lot of generalization tasks in flies. I contacted John Krakauer right away and he answered immediately. He seems to be a very nice guy (not the usual arrogance invertebrate neuroscientists get from vertebrate researchers usually) and we decided to meet up at the upcoming Neuroscience conference in Atlanta next month. I love the internet - how did we ever survive without it?