Well... just get the 8 voice version, and buy the upgrade later when you have the money... that's one good thing about the REV2, that it's expandable.
Regarding having 16 voices in comparison to 8, that is about it really... the polyphony... personally I would not be able to do with just 8 voices, simply because the joint forces of two layers is a lot more flexible than you'd expect really... especially if the sounds you want to use demands chord playing, or have long release times... it's really when those two layers combine to form a sound, that you can do so much more with the REV2, especially when you have 16 voices available.
But other than the polyphony, there are no differences... the only one I can think of is if someone made a program on a 16-voice that use unison with full amount of voices allocated... then it would not sound the same... a program can be programmed to use a single layer, with unison set to 16 voices... if you play that on an 8'voice, the REV2 of course automatically makes that 8 voices... samt thing in stacked or split mode... if it's programmed to use more than 4 voices each, then an 8-voice REV2 would not play the program like it was meant to be played.