Indeed, my solution for (2) is to say "hey, let's make a real set of ViewModel objects". I have an internal-only project that I built myself, which was basically a fairly complex experiment in using Knockout - and making the ViewModel objects myself really made a difference in how the whole application felt, it felt great. But it was a hard sell - one that I lost - when I tried to sell them on it in general.
And indeed, I don't know if I'll ever find a solution around (5). I feel like it's a small price to pay for the benefits personally, but it is a price to be paid. :(
Yes, it is worth the effort to check out. While you can mash several things like sammy and other frameworks to make your own SPA type solution, Durandal already has done it for you (built on require.js, sammy, and knockout)
3)I found binding handlers in the beginning to be more work to write, but in the long run the code is less brittle and much more reusable (not tied to a specific element). Although pitching to front end folks may not care about this.
4) If I'm understanding what you are saying this sounds to me like you are hitting the limits of Knockout. Durandal has a much better structured system for models/views and nested bindings. The compose syntax in Durandal also comparable to the template binding in knockout, although a lot more flexible. Durandal is built on knockout so much of what you already know will just work.
5) Agreed