Vacuum chuck plus z map seems like it would be the best trick.
I have Renishaw probes on my machine which is technically capable of pulling that off. The reality is that I would have to write custom software on the machine and on my desktop to pull it off. It may be a fun challenge, but the reality is the time necessary would be hard to find.
Regarding the high frequency stuff: I don't know what the surface finish is that milling leaves behind but it might result in higher loss than expected. For projects I am working on (mind you, these are > 100 GHz things so I don't know how this scales to 20 GHz) surface roughness of the materials is a big deal and causes a lot of additional losses. You might find the same happens when the edges are rough due to milling?
I'm curious as to the performance here and would be interested to be able to compare some results.
The edge finish is a combination of the tool, spindle speed, linear feed rate, and total runout at the tool tip. The Chemical process almost certainly has smoother edges but not necessarily consistent. At 20Ghz - things are rather sensitive. At 100Ghz - everything is broken and you just manage various levels of broken. Not sure how you pull that off - you are decades ahead of me.
I believe I can pull off very clean and consistent edges as far as machining is concerned, but not sure how it would compare to a normal chemical processed PCB. Keep in mind, my goal is to use this process to -
A: Do practical experiments to better understand the engineering and scientific principles as I study them.
B: So sanity checks on specific design elements to avoid a bust on expensive, special impedance controlled, fancy substrate, blind/buried vias, PCB prototypes.
C: Make various 1-off test fixtures
I do not expect to re-invent the wheel on the bleeding edge of RF design - only use it to educate myself on how variations in geometry impact the signals and how simulation varies from reality. Driven by 50% business needs and 50% pure curiosity.