There's no cost to privacy if it's all being written to /dev/null. I'm not worried because the cost benefit analysis is not even remotely in the video conferencing app's favor to listen to that traffic. Are they really going to use the compute time to analyze all this audio, then do what? Try and monetize data on what people are doing in the background of their video calls?

The technical cost of deploying this is probably large, and the cost to reputation immense if they were caught doing this. By comparison, giving people the additional sense of privacy by actually turning off and on the mic is likely more than outweighed by the annoyance of lag between turning on and off your mic and being heard by the other chat members.

Although they could do something like write random bits of audio to the stream when the mic is muted in software. That'd at least let users know that the actual audio isn't leaving their device. But the hardware peripheral activation is probably not going to go away.