The file mentioned in the README is compressed and, given the I/O vs compute balance of modern computers, most would consider it a best practice to always operate on compressed data, so it seems like the benchmark should recognize that and require decompression of the input.

Yes, that makes it less trivial, but it's much more real world. It would also test the quality of the gzip/bgzip library implementation for the language. Perhaps report numbers for both compressed and uncompressed inputs if you want to isolate the compression effects.