Keep in mind, not everyone with Soc11+ is automatically a Noble (has a title, fief, all that stuff); they may just be very well-known and very well-connected.
Even being "born" with Soc 11+ doesn't have to mean that that the character is automatically of a Noble family; they may be the child of a megacorporation executive.
Actually, Soc 11+ DOES mean they have a Patent of Nobility, which is the defining trait.
The ones who have feifs, they are rare. 11000 worlds, roughly 11000 feifs... out of some 300,000,000,000+ persons... 11000 of roughly (very roughly) 15,000,000,000 nobles rolled. I think TNE mad a decent enough correction by using 2d6-1 for soc.
I do agree that the character generation is not reflective of the body of the population; I do not agree that it is divorced from the realities of the setting entirely. I'd say that Soc should be generated differently, or nobility should be handled differently.
Given a typical dark ages construct, nobles and gentry combined account for about 10-15 persons per 300-500, or somewhat between 2% and 5% of the population... (1 knight and his family on a manor supported by 300-500 peasants in several villages.)
I would expect lower end numbers for a tech society. The UK has what, a few thousand of many millions?
I'd agree that the input to CG is broken (That is, the 18yo characters themselves are not reflective of the imperial population's 18yo's), but that the rest of the mechanic is correct
as far as it goes. That is, the skills and abilities gains are reasonable, and that surviving a term unscathed is represented well enough by the tables...
Certain elements are decent enough for resolving applicants (commission, promotion), but neither is required and thus the intangible "do I want to apply" renders pure statistical analysis impossible.
I find it to model "Citizens of the Imperium", that is, persons who have been in imperial service and/or off-world work, probably quite well... but is unlikely to represent the real "norms" of locals.
But it also is an excellent mine for implying setting issues.