the experience i'm having is that the loop() process seems to run at a fixed rate per compile, but that rate varies depending on the content. i'm even finding that adding an extra multiply tips at a place and gives me a slower loop. but, functioning at a steady rate.
so, for my implementation, i will have to find equilibrium. the first simple sketch gave me an incredible samplerate around 300k, but with a procedural voice synthesis engine it went down and down per compile, i'm currently fighting about 8000 samplerate.. i'm on a border of my performance tolerance here so this trinket m0 is pulling just about enough for three formant bands, but i can probably find an optimisation. 4k nyquist is pretty cheesy.