Let's see

1) no, my input device is an optical stream with varying sample rate and I was hoping the convert device would magically adapt to the different rates without Camilla config changes - it didn't.

2) the load increase is negligible. the total is <<20% usage of my pi 4.

3) the delay could easily change, but the important thing (that the two delays are roughly linked and near each other) I expect is true for all sample rates. I was hoping to test more stuff and with more quantitative data but I didn't really have an obvious way to do that.