So, after more testing I found that a lower value resistor is needed in the bias circuit. Using the 100 ohm 10W resistor there was too much voltage sag on speech peaks, which causes the amp to run into class C territory. I decided that halving the resistance should do it, so I took out the 100 ohm, popped in a 50 ohm and took some voltage readings at the bases of the transistors: .650V under full load, .685 at idle. I also noticed a slight increase in power output, about 5% across the frequency range, and the transistors seemed to run just slightly hotter running into an MFJ dummy load. Everything still looks just fine on the scope. So far these are promising results..
After replacing the resistor, the transistor switch that was described earlier went from optional to a necessity. The 50 ohm resistor heated up more than I wanted when the bias circuit was allowed to run all the time, so below you'll find a pic of the solution for that.
The transistor used is a TIP42G PNP power transistor. You don't necessarily need a power transistor, a small-signal PNP will do the job in most cases, but since this mod relies on thick component leads for structural support to avoid needing a PCB it seemed ideal. The resistor that the wire from the relay diode connects to is 220 ohms. The 220 ohm resistor is connected to the base pin (right in pic) of the TIP42G transistor. The transistor's emitter pin (leftmost in pic) is connected to the top pin on the front panel switch as in the picture. The last (middle) pin on the TIP42G is connected to the 50 ohm resistor in the bias circuit itself.
*It's VERY important to ensure that everything is positioned so that the tab on the TIP42G cannot touch ground, so avoiding anything metal is the best way to insure this..