Re: Biggest issue for Intel
I don't see producing an LLVM backend as being any more work for Google than the effort of producing cross compilers and binary blobs for MIPS, various ARM architectures (e.g. v6, v7, NEON fp support etc.). It would make the NDK smaller and more maintainable and obviously lessen the burden on developers who use it too since they only have to run the compiler once. I assume the NDK could ship with an gcc-llvm compiler, the developer would distribute the bitcode with their package and the runtime would throw the bitcode through a compiler first time it was invoked and cache it somewhere.
I think Intel could have a motivation to help provide this since it would benefit them first and foremost since they are the underdogs here. Anything which gets code to work with their platform more easily is an advantage.
This would benefit not just games mind, but also any C/C++ codebase. For example LibreOffice, VLC, Firefox etc.
As for games, most of them will stick closely to OpenGL ES 1.1, 2.0 or now 3.0 as far as practicable. I know from my own experiences that certain GPUs can be fickle and some interpret the spec more loosely than others so there are still occasions where you might have to special case a GPU, or use an extension, but generally the default is use the API until a reason arises not do so.