Jon,
I would say that chemistry (of the glue) isn’t the issue, but physics..
the noise issue arises when a flexible (part of the) baseboard resonates due to the vibrations arising from the model. This resonance, like the soundboard of a musical instrument, amplifies the noise. This implies that there is a physical connection between the track and the baseboard, which could be a rigid glue, or mechanical fixings like screws or pins. If the track is glued to a flexible layer, which itself is glued to the baseboard, with no mechanical connection between track and board, the hope is that the vibrations from the model are largely absorbed in the underlay.
PVA which does dry to a fairly rigid solid is likely to transmit the vibration to the top of the underlay. It may also soak into the underlay, and thus turn that into a rigid layer which may meet the scenic objectives, but offers no sound deadening or vibration barrier. This would also be the case, of course, if the layout were ballasted, and the glue and ballast form a rigid concrete-like, vibration-conducting path to the boards.
Contact glues tend to stay as thin layers, which probably work well in this context, but are probably not ideal if ballast will be added.
Latex glues stay flexible, and probably offer the best compromise for gluing track and ballast, but only if the underlay is a closed cell foam, if it’s open cell, it will soak it up like a sponge. I doubt that cork has any significant sound deadening effect, and of course, stiff non-resonating baseboards will help too.
Another observation, Latex doesn’t turn granite ballast green, PVA does, that’s chemistry!
Hth
Simon