Wasn't heat. It was only keyed about 15 seconds. Immediately upon failure I felt the heat sink and it was still stone cold (the lab was unheated and 10 C.).
I originally thought that there had to be some other cause. Turned out to be the MOSFETs.
Yep! Dummy load was a 250W thick film mounted on an enormous heat sink and better than 35 dB return loss. Input during bias setting was BNC mount with 32 dB RL. Power supply was a well filtered and regulated analog Harrison Labs. A separate 4.5 volt supply applied PTT signal.
Looks like a little bit of DIBL to me. Not something to be expected on a big 'ol power MOSFET. But sure feels like.
The amplifier works perfectly, just exactly like it is supposed to. Up to the point of failure (which is my fault entirely) the original transistors did, too. My point in posting was that it might be prudent to perform the bias setting at the highest value of supply voltage anticipated. Can't hurt.
I would not assume that there are many amplifiers built where such large excursions of supply voltage are encountered when considering bias settings. Most of those are in linear service and not as subject to variations in threshold. This is actually the first time that I have ever blown an IRF510, come to think about it. I've used them in a lot of rigs. An incredible device and dirt cheap. Then again, this is the first time that I've pushed them this hard. My bad.