If you test small features, you get knocked for concentrating on microbenchmarks which don't have any real-world meaning. If you test whole applications or frameworks, you get criticised for benchmarking that rather than the language implementation.