intermittent failures could easily be either microwave load oscillations, or low frequency bias line oscillations. have you tried a varying load to try to induce failure--like a 3 dB pad in front of a sliding short?
You might have trouble seeing it, the oscillation can start up ~ randomly, and blow up the device before you could see it. Use spectrum analyzers on "max hold" to try to see it
Self oscillations are not present;
I've noted that current gate changes polarity when PA is close to saturation: from -0,2mA to +0.28mA but it is always a low current compared to the specified from the datasheet, however I've used a 5,6 ohm series resistor which is advised from Test-board...
I suspect thermal effect because with the same package there is a 90 W version of transistor