They explain it, but it's still deceptive.

I would have expected a honest benchmark result. Which will be much worse than 6%.

Which is totally fine! an interpreter will of course be much slower. But this seems like the wrong kind of marketing by omission to me.

Yeah, but they're just benchmarking a wrapper around gzip. It looks to be around 300 LOC and does very little.. It's a bit deceptive. A large table of benchmarks across different domains and workloads would be more honest.