I should have refined my comment on QA, I know how hard it is, but it is true that if you can separate the functionality (gameplay) from the content you could, to an extent, test them separately.
If the Demo has a finalized version of the functionality, it could be tested while the content is being generated.
Certainly, all content would need to be tested, and tweaks to the functionality might happen based on those test, so this model would work better with games whose architecture makes it easy to separate the two.
For example, a game like TrackMania, where the tracks are all built with the same “tiles” could have been tested with just a sample track while the mappers worked on the rest.
I, at least, am a very big proponent of proper encapsulation, so independent modules can be tested independently as well as in conjunction with others, hence my mindset here.
Besides, I think a good demo is when you’re offering a fully functional game that can be expanded by paying, rather than just a bunch of limited features. Think the very successful shareware model used by DooM back in the day.
(Again, I know this might not be economically feasible, and when you living depends on your game selling, you can’t be too idealistic)