Instead of standardizing something low-level to input/output audio and query the hardware, like an OS; it standardizes soooo many things and filters, including downmixers, panners, quadfilters and Convolution (for reverbs); which is where the issue is in this exploit.
The resulting complexity is huge, almost impossible to implement correctly in a cross-browser way, and a lot more code in C/C++ is written.
I do advocate simpler APIs for audio, low-level, and let JS or Wasm do the filtering, using the JS sandboxing model.
And, of course, this does not apply only to WebAudio, but I think WebAudio is a good symptom.