Currently both trunk and branch use the same repository for tests and tests' data. This leads to multiple problems: new features added to trunk require new tests, which won't pass on branch.
The most obvious (but not very optimal solution) is to branch the test repository.