Very nice post.
And Bill's comments too. They remind me of the LEO editor for Python which was always touted as having great productivity benefits if only your team would undergo the three month training required to use it properly. Not sure if it ever took off or could. But there's something nice about the idea that LEO empowered programmers could outperform the norm.
I'm convinced that there are certainly productivity improvements available to power-users, beyond anything currently dreamed of, once we step away from the assumption that "ease of use" equals "1-to-1 correspondance between functionality and UI objects".
As every nerd knows, real (interesting) productivity, comes from higher levels of abstraction. And maybe what's really important about the outliner tradition (from HyperScope to MORE / UserLand / OPML to LEO) is that it remains loyal to this notion. When you collapse a block of text and ideas down to a single-line, you are essentially abstracting away from that detail and working with the higher-level description.
OTOH, the Xerox Parc tradition of the GUI and direct manipulation, lost this core ideal. (At least as it was spread via Apple and Microsoft, although obviously you can probably do all sorts of powerful abstractions via a Smalltalk interface)
I'm pretty sure that this insight is general. The really interesting innovations beyond HyperScope are going to be new ways of giving the power-users yet more abstract ways of manipulating their information. Either by folding more of it together as complex aggrogates, or allowing large-scale cross-cutting processing. (Maybe style-sheets in Word are the only other surviving popular example.)