Mickey, thanks - I also share your opinion 100%. IMHO a decision is overdue. There should be one as soon as possible !
I'd tend to say that re-re-re-reinventing the wheel is a bad idea. The worst possible - spreading resources even further :-) IMHO it's not a library problem - GPE is LGPL IIRC. It's just trying to consolidate on something that people will believe will live on. Usually you don't work on something if you don't believe it'll succeed.
I believe the first alternative you listed is the right thing to do- go for application. Take the flavour of the day - even if it's already outdated, and port/create apps. Focus on the best ones and bring in free software replacements when it's needed (ex: a synchronisation daemon like shsync to let people use intellisync, libsl/dtm ...)
Leave the toolkit/core to trolltech - that wouldn't be very far from what KDE did: take a good library, and use it to write good apps. That's a model most people would believe in and IMHO would be willing to work with: a company taking care of some stuff, the community taking care of the rest. If there would be such a project, you can count me in! Ok I'm no good in C++, but I can do C, documentation, usability etc.
One could say it's playing catch up again. I tend to think it's not - starting from the same basis (ex: qtopia 2.1), applications would be the #1 issue, while filling up missing parts /non GPL parts would just be a side issues. As ljp said, it'd be building upon qtopia.
Yet trolltech would need to make the devel of the latests toolkit a little more public, to avoid surprising devels with strange changes when the newest version is released.