Oh, that'd be it then. I must have overlooked the bit where you mentioned the Silesia dataset and skipped straight to enwik9. With that corrected, I get similar results to yours.
One issue I did notice during testing is that lzav's compression ratio is fairly unimpressive (~42:1) on all-zero data. Surely it should be able to do better than that?
One issue I did notice during testing is that lzav's compression ratio is fairly unimpressive (~42:1) on all-zero data. Surely it should be able to do better than that?