Has any consideration been given to using a subset or derivation of LLVM bitcode a la PNaCl? I know there are significant potential downsides (e.g. according to [1], it's not actually very space efficient despite/because of being bit-oriented and having fancy abbreviation features), but it already has a canonical text encoding, it has been 'battle-tested' and has semantics by definition well suited for compilers, and using it as a base would generally avoid reinventing the wheel.
Of course, several of the people working on WebAssembly are actually from the PNaCl team. And others, like the Emscripten people, have lots of LLVM experience as well. WebAssembly is being designed while taking into account previous approaches in this field, including of course using LLVM bitcode for it.
LLVM bitcode is simply not a good fit for this:
* It is not portable.
* It is not designed to be compact.
* It is not designed to be quick to decode and start up
* It has undefined behavior
* It is not sandboxed.
* It is a moving target (see recent big changes to LLVM IR on geps and loads/stores)
Each of those can be tackled with a lot of effort. But overall, better results are possible using other approaches.
[1] https://aaltodoc.aalto.fi/handle/123456789/13468