You’re right, it is rather queer. I don’t know why Audacity passes short sounds to the effect, as if they would have the same duration as the longest one.
I guess that it simplifies the code execution.
I can only talk for Nyquist effects. If I want to write a really multi-track selection suitable plug-in, I have to trim the sound, i.e. remove all zeros (didital silence) before the sound is returned to audacity.
Since this procedure is time-consuming, it is almost never applied in a plug-in. In fact, I am only aware of my align-track plug-in, that does this (in order to align tracks end-to-end without unwanted pauses).
Aligning tracks end-to-end
However, it does this also for the beginning of the track, which is not desired for tracks that were deliberately shifted.
Fortunately, There is a command that removes the unwanted silence from your audio (after processing):
Edit > Clip Boundaries > Detach at Silence (ctrl-alt-j)
Thus, just select all as usual, apply the effect you want and execute the command above.