I can't experiment with these instruments, but I can speculate as to why gliding might begin from unplayed notes. If each oscillator keeps track of its last note played, then that last note memory location could be left undefined when the instrument is started*. If this is the case, it could be anything, since "undefined" is basically "random value."
Doing this on a per-oscillator basis is questionable anyway. If you play a three-note C# minor chord, say G#, C#, E, and then go up an octave and turn that into C# major with F, G#, C#, I'd think you'd want the G# to glide to the higher G#, the C# to glide to the higher C#, and the E to glide to the F. It shouldn't matter which oscillators the new notes are played on. That turns it into a hardware implementation issue and not a musical issue. But of course the software needs to be a lot smarter.
* Or perhaps when a patch is changed?