You guys check my work, but this seems feasible:
Assuming it's a 100 watt amp and he's got two speaker outputs, one of which is occupied by an 8 ohm load, solving P=(V^2)R for V, we get 28 volts across the load. OK, double that for a cushion to 50 volts.
If he builds a 56:1 voltage divider from a 560K resistor in series with a 1k resistor and connects it to the other (parallel) output of the amp, the amp would still see an 8 ohm load and the voltage divider would dissipate less than .005 watts, plenty low enough power to use 1/4 watt resistors. If he takes a signal from the terminals of the 1 k resistor to feed the mixer (one terminal of the 1 k resistor would have to be ground common to the amp and mixer), he should get less than 1 volt of signal, and 1 volt is the standard for 0 VU. The 1 k resistor is large enough that there should not be any impedance mismatch probs going into the mixer, and his amp would not notice the minuscule change in load.
Any reason why this would not work?