The important thing is not to inflate point costs for social engineers and face characters. That's
already a marginal role in most campaigns, and ruling that all the skills such people contribute to the group are at -3 unless they pay extra for CFs they might use just once only devalues it further (possibly to the "Screw talking, let's just kill them!" level). Yes, for the real-world definition of "culture," there are lots and lots of distinct groups.
In a roleplaying game, though, a few broad groups will do. Cultural Adaptability essentially caps the number of groups at 10 . . . and though that's a "cinematic" trait, it bears mention that total, immersive familiarity with 10+ real-world cultures is no less cinematic.
In the balance, then, I'd say there are two fair models:
- Lots of finely divided CFs. Cultural Adaptability is allowed and becomes the go-to trait for face characters, just as Combat Reflexes is for warriors and Flexibility is for sneaky cat burglars. Lack of CF becomes a not-a-social-engineer problem, and mostly irrelevant to PCs who don't do much talking anyway . . . their players won't want to pay 10 points for CA, and won't see the point of buying 1/n of the available CFs for large n.
- At most 10 broad CFs, but Cultural Adaptability is off the menu. As above, really, but now the non-face characters can meaningfully buy in if they want occasional speaking parts, because every point spent spans a solid 10% of the space.
The real world goes with "lots of finely divided CFs but Cultural Adaptability is off the menu," but that doesn't make for a fun game because it penalizes pretty much everyone.