The query we had that fetched the ISBNs for sending to NoveList was pretty fast, but NoveList needs more information. The query we have now gets all the right information, but on a real library's dataset it's unusably slow.

Ideally we would not have to send this information to NoveList. If we must send it, the best solution would be to improve the query's performance. If that's not possible, we could cache the NoveList 'representation' of an Edition, the way we cache the OPDS representation of a Work, and pull those cached representations out of a much simpler query rather than trying to construct them in SQL.