I found Sierra Circuits have an
online impedance calculator that specifically covers FPC under 'coated microstrip differential pair'. This is also what they describe FPC as in terms of impedance calculations in their blog articles.
The calculator allows you to enter different Er values for core and coverlay. If I plug in the appropriate values, and aim for an Zdiff of 85 ohms and a fairly sensible trace spacing of 0.2 mm, then it gives me a trace width of 60 um (~2.3 mils) to get actual Zdiff of 84.7 ohms. This is about what I managed to work out with Saturn PCB Toolkit using edge-coupled embedded calculations and fudging for a number between what was output for 9 and 18 um copper (62 um / 2.45 mils).
Things get much more sensible if I specify 50 um core polyimide thickness (gives trace width of 126 um / 5 mils), but I won't have that option - none of the cheaper players like JLCPCB or PCBWay offer anything other than 25 um core thickness.
If I aim lower, say 80 ohms, I can get 66 um trace width, which is more reasonable, but still < 3 mils.
However, I note that in one of Sierra's articles discussing hatched ground planes, it says:
Since the impedance of a trace across the hatch ground zone is greater than that of the solid ground region, the inductance of the trace must be reduced to keep the impedance under control. This is the reason why a trace should be built a little wider. This lowers the inductance of the trace and raises the overall capacitance with respect to the hatch ground.
(BTW, why the #$%& do they disable right-click and copy-paste on their website?! I had to break out the browser dev tools to copy it.
)
So, if I use a hatched ground plane, I should be making my traces wider anyway. But I'm not sure by how much. Maybe if I have cross-hatched ground plane pattern that reduces the area of copper to around 60% (is there a way to calculate this from hatch line width and spacing?) I could increase my trace width by, say, 30%?