A simple solution would be to duplicate the sample and assign each keyzone to a different FX chain.
Remember that FX would be applied to all samples assigned to that chain, so you need to split it up at the sample level to get ‘truly’ different FX applied at different velocities - you can’t just have the keyzone velocity mapped to some parameter, as it would still affect all sounds in that FX chain (adding a keytracking device to the FX chains to control a delay feedback would work just the same)
Related, perhaps: sample aliases?