1- As to the bypass of the source resistor I have been using a very careful simulated array of caps (not only one or two caps), so that the AC impedance to ground is less than 0.3 Ohms from 1MHz to at least 3 times the operating frequency range.
2- The electrical isolation from ground is provided by thermal conducting material. Comparing the temperature of a 180 mils flanged package for a 10Watts PA at full power after thermal balance be reached, I got 4C of difference. The reliability of the device regarding this small increase in temperature doesn't change.
3- The connection of the bypass capacitor array to the FET follows different techniques depending on the kind of package. As I said before a multi-decade bypassed has been achieved and any degradation in performance or instability has been noted in an extended temperature range.
Some additional comments:
1- A GaAs FET when pushed to a light class AB tends to increase the gate current at low temperatures(-40C range). For a well matched device for example in a 10W amplifier, it is easy to have more than 40mA flowing through the gate junction at low temperature, so that a low power inverter can be a true problem.
2- I believe in what you say about to keep the leakage of switching frequency at low level only if you have a reasonable physical distance between the VCO and the PA. However in compact wireless packages you don't have more than 2 inch between VCO and PA and so the problem of periodic noise is real, even more in narrow band transmitters. This for not to say about the space used by an inverter and associated circuit.
So, the arguments you presented above for not to use a floating source's FET are correct but surpassed.
Thanks for all the considerations but I continue to stimulate the use of a floating source bias.
NandoPG