I tested on the dickens dataset (Charles Dickens novels). This algorithm works well on English text. I do not expect it to be any better than zstd at binary data. I haven't tried code.

It does used zstd. My algorithm is a preprocessing step for zstd.