I think they forgot to include the benchmark time for "npm (cached)" inside the Binary Manifest Caching section. We have bun, bun (cached), npm. I think the summary statistics are also incorrect.

They don’t seem to clear the cache in between fresh runs. This is evident from the lower bound range time being the same time as the cached mean.

This leads them to the incorrect conclusion that bun fresh runs are faster than npm cached, which doesn’t seem to be the case.