I have been looking for this since quite some time.

I wish this would have included a "control test" of benchmarking the underlying block device directly.

F2FS results are interesting. Very inconsistent but for some patterns performing surprisingly well. Default options would imply using zstd compression, which I guess hs a major impact.