CMS's link should show you for small data sets they're all fast, the only thing i'd suggest is that you avoid for (a in b)
as it has a much higher memory overhead than any other looping construct and can potentially be much slower due to its "interesting" semantics.
Anyhoo, with the exception of for(in)
any real JS should be spending quite a bit more time actually doing real work than is spent handling the looping itself, so minor variation in cost of the loop shouldn't be too important.