Well Mark, this one has me stumped. A number of thing are disturbing me here.
1. The output power is definitely low. But I don't understand your numbers. 15.3vp-p is 5.4vrms (15.3÷2×0.7071=5.4vrms) which across an 8.2Ω load should translate to 3.6Wrms (5.4^2/8.2=3.6). This is REALLY low. Time to check the bias points.
2. Upper rolloff design point is 76kHz, 47kHz is too low.
3. Channel separation seem good.
4. The 5.6Ω across the 8Ω tap lowers the reflected load impedance to around 2.5kΩ. The band pass opened up to design points (which is good) and the power went up slightly as expected.
5. A 5.6Ω load across the 16Ω tap lowers the load to ≈1.2kΩ. When I put this load on the original load plot, I calculate about 10.5W, so this matches. But your second order harmonic distortion should have increased SUBSTANTIALLY. I don't understand this at all.
As to running your speakers on the 16Ω tap, I would recommend against it. With the reflected load the upper swing is significantly above the plate dissipation limit and played loud for an extended period, you may melt your plates.
Without having the amp in front of me I really can't offer any more insights that this. Something seems wrong with the amp's performance with the matched load on the output. Some investigation may be in order to see what other changes may have been made to the circuit.