I agree that speed is important, as I don't want to slow down users who just want to click on an ep and play it. I would very much hope that the proposed stacked window could be made to appear as quickly as the current dialogue takes to slide up, or at least make the buttons appear just as quickly and be usable immediately even if the shownotes take a couple of seconds to fill in.
That's a good point, although there's no particular reason to limit the toolbar in my proposal to just one line, so it could be made two lines or more to accommodate more options. Obviously that will make the space for the shownotes smaller, though, so there is a limit to this.

As you say, it depends on the paradigm you are going for. Personally, I see the episode as an "object" and so it makes sense to have a stacked window being part of the back/forward navigation and showing more information about the object (which could include file size, duration etc., dates published/downloaded etc. within the shownotes scrollable area) with the actions available below.

Another idea I had which I mentioned in the https://bugs.maemo.org/11501 report is that once an episode is represented as a window rather than a dialogue, there is the possibility of adding extra functions. For example, a swipe left/right (and/or hardware keyboard cursor left/right) within the window could move to the previous/next episode in the list, which could be useful for flicking through eps to find a particular one. This isn't possible with the dialogue implementation.