That said, I agree that the __dict__ proposal is a hack, but as is the current _removetestAtIndex mechanism.

The only clean solution I can think of would be to have two separate classes:
- a TestSpec which contains execution-independent data about a test case, and knows how to instantiate it
- a TestCase that is used for the actual test execution, but isn't saved in the test suite

Maybe it's possible to do this without any backwards compat problem by making TestSuite.__iter__ always return TestCases (but freshly-created ones, from the inner test specs). The main point of adaptation would be TestLoader.loadTestsFromTestCase().