The plugins are from different "quality". The problem is that the sound creation runs at
s-rate (sample time) and the parameter changes are running at
k-rate (control time). In the same time window, much lesser control changes (reading value, working with that value) are done than sound calculations. This can produce sharp steps when changing a parameter (especcially when changing it fast) and this can be heared.
You can avoid this by trying to interpolate values between
k-rate steps. But this means to put more code into a plugin. I havn't done this for Dexed - even for the volume I havn't done this because I use Dexed without realtime parameter change - but it is not impossible to implement