The issue here is that NaCl and Xax are proprietary approaches done inside companies coupled very tightly with existing ISAs and not developed entirely in the open.
They are both brittle with their own particular weaknesses - NaCl is weak at handling dynamically generated code, and Xax has safety issues (there are user mode instruction sequences that can freeze some x86).
PCC may be clever, but the fact remains less effort would be required to add sandboxing of processes and peons and yield higher performance.
PNaCl from what I recall relies on LLVM which is yet to demonstrate low latency code generation performance - in contrast to older more mature code generators such as TAOS, LuaJit.
Don't get me wrong, I am in favour of a WebCPU component for low latency, near native code emission and execution. I just don't think those with the pedigree are doing the work.