1.
You are right. 25 Ohm TL impedance is required to built almost perfect 1:4 impedance transformer.
However, the target Zload of the device is not real 12.5 Ohm but complex impedance. Using 35 Ohm coax provides additional reactance at the operating frequencies. It compensates reactance of the target Zload. I simulated load circuit of this design for 25 and 35 Ohm coaxes (DX: 66 vs 84).
2.
The problem of the load matching design is that the optimal load impedance of the transistor is a function of the output power. Usually, the load circuit is designed to maximize the gain at target maximum power (P1dB or P3dB). At different power level this circuit won't be the best match and sequently the gain will depend on the output power. At the target maximum power level, the gain will achieve it's maximum possible value.
In general, one could optimize load circuit at different power level. For example, if you don't need maximum power from 10W device, an amplifier can be optimized to provide maximum gain at 3W output. The gain vs power curve would have max at 3W.
In typical narrow band class AB amplifier design, the load circuit is optimized at maximum power. At low quiescent current, the gain will drop at low power. If the current increases, the low power gain increases as well. The gain vs output power curve becomes almost flat if the quiescent current is about 10% of the peak transistor current.
Optimal quiescent current is always a compromise between amplifier power efficiency and linearity.
Estimated A+B current for MRFX600H is 2*[600W/62V]*10% = 1.9A. This current value will cause noticeable quiescent power dissipation of ~ 120W. Our 230MH narrowband design operates at lower bias current. The gain curve drops at low power. You can see typical curves on page 4 of the board description:
3.
In a broadband design, like NXP MRFX600H 87.5-108 MHz reference circuit, no matching circuit exist that could optimize impedance in the whole operating frequency band and different power levels. In this amplifier, the design target was to achieve flat gain and power performance vs frequency at fixed input power. Generally, the matching is not the best at any given combination of the frequency and power, but, it is nearby optimal.
If the matching is not the best, increasing the power level can move load impedance of the transistor in the direction from or toward the impedance of the loadg circuit and thus the gain can change in either direction. It is the reason why I can not provide a general rule for quiescent current to achieve the best linearity. Optimal value should be obtained by RF testing or by simulation.
4. Please see recommended temperature compensation circuit in this application note:
https://www.nxp.com/docs/en/application-note/AN1643.pdf
