The point of the impedance selector is impedance matching. Power vs impedance is a bell curve and when impedance of your source matches impedance of your load, that is when you have the greatest power transfer to load (and, consequently, the least power lost). This is the point where your system is operating at optimum conditions.

To the best of my knowledge, the curve is essentially symmetrical on either side of the peak. That would mean if you had 6 ohm speakers, I don't think it matters whether you choose four or eight for the impedance selector. The power transfer should be the same in either case.

Now that I've opened my mouth, if anybody sees something glaringly wrong with this explanation, (or has a better one) feel free to straighten me out.