Statistical unit tests fail too often

Let's mark them as "expected to fail" so that failures are reported but do not fail the pipeline.