The verifiability policy, as it is currently called, focuses on discrete "chunks" of content. The basic idea is that it should be possible for any reader to find the material in an extant reliable source. I'll call this the chunky level. The neutral point of view (NPOV) policy, on the other hand, operates at a higher level: it is concerned with what is done with these verifiable chunks of material, how they are put together. The core concept there is that, looking at the final article, all significant views on a subject should be presented fairly, in accordance with their prevalence (that is, not giving undue weight to any given view). The neutral-ness of individual chunks isn't important, rather the overall impression. I'll call this the smooth level.
The prohibition on original research sits somewhere in between these two in terms of the way in which it operates. It applies to individual "chunks" of content, in that each must not be original thought, but it also works on a broader level by prohibiting original research by synthesis. It's not a small picture or big picture thing: it's everywhere, at every level, from every angle.
There is undoubtedly some overlap between the policies on verifiability, NPOV and no original research, but I don't think that that's inherently a problem, nor do I think that when it becomes a problem that problem can be solved by merging the policies, because the policies operate in different ways.
Merging the NPOV and no original research policies, say, would lessen the force of the prohibition on "chunk-style" original research by focusing on the overall picture painted by the chunks when put together. Similarly, merging the verifiability and no original research policies - the thrust of the attribution proposal - lessens the force of the prohibition on original research by synthesis by focusing on the "chunk" level and not in the way that the chunks are used.
I think that the best way forward would be to merge elements of the policies and guidelines on sourcing into the verifiability policy, and rename that the "attribution" policy (the name "verifiability" is often misunderstood), and to maintain the other core content policies separately. Naturally, where overlap or bloat becomes a significant problem, then the policies need to be trimmed, and I think this is where efforts need to focus from now on.