Actually my intension to create this thread is inline with your first paragraph (That's hw I got my doubt on Binary keyword)
And your 2nd para yess..I gree with them bcs if I have to share data model then,we have also done the same/followed the same.
Yess,some times 4 layer, seen it when even subsequent QVD's r also shared and even latr data model as well.
Then,fr stand alone QV-applications, I think going with only 2 layer is good only (only if Binary is just key word)..And 3 layer is definitely preferred if Binary lightens the file or make it more efficient and tht doubt remains as doubt as we need to know logics behing Binary I guess
For me it's easier (to try) to apply the same logic to all of my tasks even if they are quite simple and could be implemented with less layers to keep the number of exceptions as small as possible (especially in the beginning of the environment life-cycle).
In real life there are coming enough requirements to change and adjust things and implement new ones and often are they not well elaborate from the business and quite experimental solutions are going live without the layer-logic and other best practice approaches and I swear they won't never be cleaned properly or documented in any way ... as far as they are running but they are a nightmare if I must maintain them later ...