I vaguely remember an entry to a compression-benchmark that gamed the benchmark by treating the filename as part of the input to the decompression-algorithm, thus beating the metric that only measured the size of the file.