Overfitting on test data absolutely does mean that the model would perform better in benchmarks than it would in real life use cases.