It isn't necessarily an either-or. The gcc developers have openly contemplated switching to Clang. It is so far superior to gcc's C language frontends that there's really no contest. This would be a big win for everyone. We need similar projects for other languages.
Dragonegg could become an important player if distributions start picking it up. It would allow some of the best parts of LLVM to be integrated into gcc in a straightforward manner.
What's unfortunate is all of the license silliness going on. libc++ would have been better implemented as part of libstdc++ but Apple wasn't going to let that happen. A *BSD switch to LLVM is an unnecessary disruption if done primarily for licensing reasons.
A strong motivation for LLVM is its modularity. LLDB promises to be a tremendous step forward for free software if people can get past the GPL-only blinders. LLVM's MC project could do the same, though that's less clear to me. I certainly can imagine cool applications of it.
As for development community, LLVM has enough people behind it to ensure progress if Apple decides to drop out. But that's not going to happen any time soon. We will probably see more closed development from Apple going forward, however. I think there's a big decision coming for LLVM about how to structure governance. It works "ok" as is but even now some cracks are showing. If/when Apple does move to more closed development I think we're going see some tension and hopefully resolution of various questions and concerns.