Taking up on the thought formulated a few lines above, (you can increase your innovativeness and decrease your frustration by shortening the cycle time of any trial iteration scheme when building complex optiSLang setups and shifting yourself into a bold and curious mindset), what else can we do?
Here are a few suggestions:
Two node setup with separate output collector node (as stated previously or see the Quick Start Guide).
Setup work with a short-runner version of the AEDT project, for productive run switching back to the AEDT project settings yielding accurate results (blocking mesh refinement passes, using a coarse mesh, reduced number of frequency axis interpolations, relaxed convergence thresholds, and so on).
Going from simple to complex optiSLang projects in incremental steps, test-running as often as possible.
Using a tool for watching tree structures of master and child processes on your system, what is going on under the optiSLang umbrella?
Bring the onion shells of nested programs to consciousness: solver, AEDT, integration plugin, optiSLang system, algorithm, optiSLang project.
Try first feasibility with exemplary dummy responses before investing effort in complex calculator works for derived responses.
Try first with only input integration active (no AEDT call) before switching the solver function on.
Use calculator or Python nodes as dummies for computation-heavy or slow nodes.