Speculation here, but I thought there was an outside chance that the '100% wet sum-to-zero fix' in this new beta might have contributed. I wouldn't have even mentioned it if the difference wasn't so dramatic (especially in stereo S&H-type filter presets of mine). I'm running stereo nearly end-to-end, so I'll check the cabling / effects chain again for inverted phase issues.
I appreciate your looking into that Speed / Restart anomaly. It's no deal-breaker, but I ran into it while tapping into the HarPeggiator's deep potential for slicer-type presets. Rate variations multiply those kind of presets into multi-presets, and sync-to-MIDI ensures that the preset has a reliable point to return to.
I wanted to report that those two LEARN fixes are holding steady here. I haven't seen any drifting Key changes like I did in 3.0.0[19]. Once in a while, I'll see the uppermost right 'dot' light up in the display after a change has been made in a Diatonic / QuadraVox preset. But it doesn't persist for very long (as if it were an intentional indicator that the edit buffer differs from the saved values).
One undocumented nuance happens in Bank mode; using the encoder knob. A preset name will normally scroll, then reach a steady state. Changing the algorithm refreshes the display for a couple of seconds with the new algo, then back to the steady-state preset name.
But when the encoder is switched to either Diatonic or QuadraVox, the behavior changes slightly. In those cases, it shifts to preset name – algorithm (2 seconds) – preset name – current Key/Scale (2 seconds) – preset name. Kind of cool.
It's one of those situations where if you like it – it's a feature. If you don't like it, then it must be a bug. 
