I have provided a fix and a test in  and issued a pull request to . I have verified the fix works under Virgo ().
The fix is along the lines recommended by Arno Unkrig in the referenced thread, but I am a little concerned that Janino is not thread safe when used in this way. I have raised this concern on the janino-user mailing list (). However, this is a Janino issue. The only relevance is that the proposed fix to this issue potentially exposes logback to the thread safety issue. So it may be worth deferring pulling this fix in until we are sure Janino is suitably thread safe.
As for the test, it could be made more compact with the use of a suitable mocking library, but I wasn't sure which one to choose, so I did without. Also, it avoids creating a custom class loader to test that Janino's "parent" class loader has been correctly set. It does this by reflectively accessing a private field from Janino. Clearly this test is likely to break if Janino changes that field, but at least the current test is small and easily understood.