I think it solves a different problem. One of Brotli's strengths was a built-in dictionary tuned for the web contents, which makes it especially suitable for short data with no known priors. The compressed dictionary transport is same but tuned for each website. In comparison, rsync algorithm would make sense only when the data is long enough (say, more than 1 MB), in which case a custom dictionary wouldn't help much because the compressor already has a lot of patterns to working with. Ideally both can be implemented, where the HTTP Range request can be updated to support rsync-like algorithms.