To make it clear, I'm talking about result, what can be gotten from API action=parse.
So - how crazy the idea is? I believe it would resolve some problems, that we currently have (with only wikitext dumps), e.g. incomplete maintaince categories.
Dumps of the HTML content that is produced by parsoid are in the works; would this cover your needs?
I'm going to merge this into another task for dumps of HTML produced from expanded wikitext.