As I understand it, the main question posed by http://bibframe.org/documentation/bibframe-authority/ is "Do we need a "lightweight abstraction layer""?

Equivalently, as I understand it, the question is "Can we attach the authority-related attributes directly to a person / document / place or do we need some separate resource for them?"

My answer is (a) yes we do need a separate resource for them and (b) we can do that very easily.

(a) yes we do need a separate resource for them
For Instances where the carrier is digital and we have a URL the instance, we already need to make a differentiation between our rdfs:Resource for metadata about the instance and the rdfs:Resource that is the instance. Otherwise we can’t differentiate, for example, between the HTML version of the FRBR report and it’s HTML catalog entry. Once we've made the distinction for digital Instances, having a different genre of 'thing' for our other entities in our model is a recipe for confusion.

(b) we can do that very easily
In the foaf vocabulary there is a foaf:Person, a foaf:Document and foaf:primaryTopic which are what they say on the tin and have, I believe, exactly the semantics we’re looking for. foaf:Agent and geo:Point are obvious contenders for Organisation and Place.  We can reuse them directly or via semantic sugar (a la madsrdf:hasCloseExternalAuthority). There are other namespaces, with coverage of these areas; pretty much any of them is better than rolling our own. foaf also has the benefit of already being used by VIAF (and I believe it works for them, I've certainly not seen any complaints). 

[I was going to use some viaf.org examples, but it’s down as I write this...]

cheers
stuart