But a professional speaker design these days shouldn't require an NFS for every iteration for things like the "multiple iterations of tweak / re-test / tweak / re-test cycles" you describe.
After the design parameters are set, you start with a simulation of the speaker. Depending on the level of detail of the simulation, this can even already show possible cabinet and BR port resonances and you can try out different variants directly on the computer.
The simulation provides at least a complete spinorama for each driver of the loudspeaker. This allows you to simulate the finished LS in advance in XO programs such as VCAD. If the simulation provides realistic electrical data (e.g. impedance, excursion,...) in addition to the acoustic data, you can even determine the possible component costs for a passive version in advance.
When the first prototype is built, then one can already track down most resonances via near-field and impedance measurements (done in a few minutes) and minimize them by iterative changes.
For the optimized prototype, a complete spinorama is then created for each individual driver (plus impedance measurements) and the final XO can then be created in the XO program, or the final version can be created iteratively through XO changes in XO simulation and listening tests of the real speaker with the different XO versions.
For the final speaker version it is enough to measure the on-axis frequency response, additionally the on-axis FR of the single drivers and if these matches with the XO simulation, the loudspeaker is ready.
If you want to be absolutely sure, you can also create a complete spinorama of the final LS (but if the XO simulation program is good, it is not really necessary) - If you are in possession of a Klippel NFS, you do this, of course
So if the development of the loudspeaker is well planned, for an X-way loudspeaker, you need X + 1 NFS cycles for the complete development.