How well does this work for massive codebases and thousands of e2e tests?