If the internal APIs of the Linux kernel were boiled down to WebAssembly, then many kernel modules and drivers could be written in any language and cheaply isolated, making it harder, maybe impossible to corrupt kernel memory. IMHO that would lead to a more modular, robust kernel with lots of options for programming it, without resorting to a full microkernel design that has costly IPC.
With Wasm you could use any fully-featured programming language that compiles to Wasm, rather than severely-restricted and stylized C with a single toolchain that can generate BPF bytecode. Wasm has a formal specification and many high-performance implementations with near-native performance. BPF predated Wasm by a lot, yet not being a general purpose bytecode, hasn't had the engineering time invested that Wasm has.
from my limited understanding bpf is not general propose by design, like disallowing infinite loops etc. otherwise people have been designing competent bytecode/vm since 90's.
wasm kernel extension does sounds nice, especially after the meltdown and spectre.
