Actually if the structure on both sides in the BizTalk mapper is the same, there is no need for functoids when mapping choice blocks. In my case the mapper handles choice-blocks in repeating records just fine.
I think the problem i mentioed occurs when the structure is a choice block on the one side, and is to be mapped to a structure with some kind of abstraction on the other side. Then it can be necessary using functoids, or if it gets to advanced it may be better using a deside shape with xpath logic in a orchestration, to make some kind of work around.