In the documentation, I do spell out the caveats of the performance claims. In particular, aehobak is intended to be part of a larger storage system. Splitting into blocks and matching similar blocks are external concerns. See rsync, git and vector databases for related techniques.
The last release was not optimal for highly repetitive inputs. This is now addressed by building the suffix array blocks in reverse and referencing the ranks of the adjacent block in comparisons. The block size is now halved so that the current block and ranks both fit in L1 cache. This yields 5.4 times the speed of the bsdiff crate.
https://crates.io/crates/aehobak/0.0.15
If you have a fediverse account, you can quote this note from your own instance. Search https://floss.social/users/barrbrain/statuses/115493741693091195 on your instance and quote it. (Note that quoting is not supported in Mastodon.)