Hi Dirty
I haven't a Snow to test but on paper specifications TI|1 and TI|2 have >80 and >100 voices respectively whereas Snow has >50 voices.
As it is "upto" 50 voices I'm guessing this is only when simple patches are being used. As with the TI's, the Snow's polyphony may be reduced substantially when either Unison, the Moog analogue filter, or DSP hungry oscillators are being used (such as Wavetable PWM, Complex Graintable, Complex Formant).
So it's kinda "how long is a piece of string" depending on how DSP hungry the actual individual patches are. You mention Bass + Pad + Arp... Bass patches are usually quite simple, Pads usually are played in chords and may use unison or more complex oscillators and will stand to eat most of the polyphony (also bear in mind when you change chords extra polyphony headroom is needed if you don't want the release of the last chord to be cut off [also known as 'note stealing']), the arp may use unison too.
Would be nice to hear from Snow users how they find things work out in reality.