Is the 85MB test file available? I wasn't able to find it. I'd like to test this result against nerve [0], a brainfuck to x86_64 asm compiler I wrote (like funkicrab, it is also written in Rust).
bff [1] has no fancy assembly, not even a compiler, just some basic loop unrolling and jump precomputation, but I too wonder how it compares to wc.
[1] https://github.com/apankrat/bff/
-- edit --
Found words.txt with 466551 words, so replicating it 28 times gets it to the OP's test size. Bff is ~5x slower than 'wc -w' ... which is not unexpected, but still slightly disappointing.
https://github.com/dwyl/english-words/blob/master/words.txt?...
19.55 user 21.49 system 0:41.13 elapsed
However, you're using 8-bit cells, so the resulting word count is incorrect. In case you're wondering, the file was generated with : python3 -m faker -o words.txt text -r [large number] -s '\n'