If the delay is really only from some kind of joint effect when all the terms from 100 lines of code are present, it seems unlikely this will ever really get solved unless you or someone else digs into the internals with the full code. However, I would guess that either portion of it counts for a lot of the slowdown, or else a bunch of pieces each contribute roughly the same order of magnitude slowdown and the 10s slowdown is the sum of all of them. In either case, there would be something small you could share, e.g. “each of these pieces takes 20 ms and I’ve got a zillion of them so overall it adds 10s”, or “this piece takes 8s when some parameter n = 10000”, or whatever.
I get it’s frustrating that it’s not just already fast, or there isn’t some quick tweak to make, but it seems like that’s the scenario here, and if you don’t want to share something so that other folks can try to debug and improve the software, it’ll probably never be fast for this (unless someone else comes along with a similar problem and figures out what bit is slow).