For some reason the factor "2:1" came into my head. Once you try to compress sound data by more than this amount, the compression starts to impact upon sound quality.

Clearly the overall sound of a "sample based" keyboard depends on a number of factors: the original samples (quality / length), the compression method, the internal programming of the keyboard, applied effects, etc etc etc.

I always find it interesting (and a good indicator of basic quality) to listen to sounds as single voices with all effects turned off.