I don't see anyone tracking down what PCM format was written to pulse vs what format the pulse daemon had to write to the driver to determine if format conversion was necessary. Needless to say, the gstreamer->pulse support has changed greatly since 2008.

But as it were, profiling my system (Fedora 30) with Sysprof to look at where time is spent:

  - Roughly about 2% CPU amortized across all cores at idle, but playing an MP3 in rythmbox (so includes shell, rhythmbox, a webbrowser, etc)
  - Of that 2%, 7% (so .14%) of that time was spent in the pulse daemon
  - .04% (2% of 2%) was spent in libspeexdsp.so and all other samples were negligible

If you're still having problems today on your system, I highly suggest running sysprof (or perf even) to get a quick high-level overview of where time is spent.

It's practically one comment each year in the last five years, doesn't look like a widespread problem.

I've had the issue (Ubuntu with VM). I think I switched to ALSA, or just disabled sound. On Linux when I "fix" a problem I very rarely write a comment on a forum...