Fortunately current btcrecover developer already provide such address database, although it was last updated on 2022. Although it shouldn't be problem since OP said "A few years ago...".
Do you know how the btcrecover devs "refresh" the address database? What kind of tools do they use?
btcrecover documentation mention 2 ways to create the address DB (from blk*dat file[1] and list of addresses[1]). But the developer doesn't mention which method he use to create his prepared DB. On the documentation, he mention few .py file script.
Isn't 115 minutes (at best 1.1k permutation/sec) too slow? Although FinderOuter doesn't have this feature but I get 22k perm/sec for BIP39 and 1.2m perm/sec for Electrum mnemonic on CPU when recovering 12-word mnemonics with missing words.
Nice catch, reference benchmark shows i7-8750 shows it has 33K permutation/sec[3]. So it's possible OP used sub optimal configuration.