With a transistor that had more gain than desired mounted to the heatsink, he applied just enough forward DC bias to begin scrubbing some gain off.
Sounds to me as if he was blowing out individual emitter-resistor wires with the overload current. That was the structural breakthrough that made high-power RF transistors practical. A single chunk of silicon large enough for high RF power won't heat evenly. The part that gets hot first now has a higher current gain than the cooler region next to it. The hotter region with higher gain will 'hog' circuit current away from the surronding less-hot area. Until a hole is melted in that region of the die. RIP transistor.
Fabbing 8 or a dozen parallel 1-Amp transistors in parallel on one silicon die gets you an 8 or 10 Amp equivalent transistor inside a single package. This allows the use of a thin nichrome (I think) wire in the emitter of each transistor. If you have ever looked inside a big audio amp or linear-regulated power supply with eight or more output transistors, there's a small power resistor alongside each one. Same principle. The transistor that heats up first draws the most current through that resistor. This reduces the base bias on that one, throwing more of the load onto the cooler transistors. Serves as negative thermal feedback, sorta.
When you put transistors in parallel this has the result of increasing current gain, or 'hfe', or 'beta' over what a single one shows.
Seems to reason that popping those tiny nichrome wires one or two at a time would reduce the measured current gain. I'm not a solid-state physics expert, but this is the only way I can explain his trick.
OTOH, crippling a transistor that used to have twelve parallel circuits down to ten or eleven of them can't help the RF-power gain. Even if it improved component matching at the low current of a tester, it should downgrade how the transistors share the load at peak currents. If one part has all twelve transistors going, and the one alongside has only ten remaining after it's "matched", I can't see what's good about that. Who cares if they match nicely at idle current with no drive, but won't share the load evenly at peak power?
I would rather have transistors balanced at the high-power end of the curve. That's when they get hot and the balance matters.
73
