The trick would be to get all of that goal-search outcome
set, into a form that the optimizer can deal with.
An alternative is for you to do the same variations on
the circuit, for N different analyses, and collate / inspect
for yourself - 3D plots galore, or whatever analysis you
can make and prefer.
I've never trusted, so never cared about, "canned"
optimization. But I've sure shotgunned many circuits
looking for gradients in performance attributes, to
go sledding.
If you break up the simulations rather than making
them inhabit some larger, single bundle it'd be easy to
farm them out to separate servers and get a lot better
run-times (and, probably ease the data reduction as
analyses will stand by themselves in regular format,
not be a mash-up that needs separated anyhow).