Any work I have done has focused solely on getting *something* to work. I've essentially tested if NSString works, and that's about it.
The best way to determine whether something you're interested works is to try it out. I wasn't personally interested in getting more than a proof that it is possible to continue to work on gnustep-base. And it is possible. One can sit down, iterate and contribute upstream. Even that's fantastic news compared to, say, four years ago. You "only" have to deal with a nonstandard libc, nonstandard directory layouts, nonstandard sandboxing approaches and a requirement to interface with Java. You don't have to build your own compiler, deal with a custom build system or (as was the case until recently on certain other platforms) find a way to target a proprietary bytecode VM...
I'd say that, similar to any other platform where portability is not proven yet nor guaranteed, it is best to start a project 'from scratch' and work on target ports 'in parallel', adapting to quirks of each in the process. Based on what experience I had with targeting Android in general (and that isn't much), I'd treat porting existing code akin to starting a new port of a game that draws with DirectX and in general shows little consideration for non-Win32 platforms -- one year into the project. It's probably not going to end well (or at least not without many stolen staplers and burning buildings).