You make some excellent points, especially the one about banks and molasses. Scraping websites in general is subject to so much arbitrary change you tend to give up after a while.
With reference to the rant about importing data directly into one's accounting software of choice, the random nature of bank data formats is more likely to render it impossible, especially for the application developers who I'd say would be weighed down with maintaining the core application.
I think the best option would be for the core accounting application to have a well defined format for import/export, leaving the work of getting data into that format for another application or project like the ones mentioned in other sub-threads here (simplefin.org looks interesting aready).
I currently use a FOSS (I think) Windows application called TurboCash (turbocash.net) which does a pretty good job, but also requires some working to get bank data in. I'm thinking of trying out ledger-cli, with some hack work on getting bank data in. With my recent drive toward the command line, I'm hoping this may be my accounting Valhalla.