The main problem (in my particular case) is that single test cases are slow (taking up to one or two seconds on my slow machine). Splitting the test file into smaller chunks (How many tests per file? 30? 20? 10?) would still leave me the problem that it could take 10 seconds or more to actually run the faulty test --- to long for development.

And for the "too complicated" argument: the failure of a test is a unusual thing, and normally only I will see the failure and very probably it's also just me who will debug and handle the error. A solution which works for me is absolutely OK. I don't need a "nice" solution, just a working solution.