If the change in power level is accomplished by decreasing or increasing plate voltage, the impedance shouldn't change. If it is modulating linearly, any change in plate voltage should result in a proportional change in plate current. The change in RF output power should be directly proportional to the change in DC input power to the plate of the final. Think of it this way. When you plate modulate a final, the plate (and screen) voltage are changing all the time, from zero to twice or more the resting plate voltage, throughout the audio cycle. If the impedance changed with plate voltage (and thus power level), the impedance would be continually varying in step with the audio as the transmitter is modulated. You would have "impedance modulation" of the final as well as plate modulation. This would undoubtedly result in undesired phase modulation and other forms of distortion, and would indicate severe modulation non-linearity. The final amplifier impedance should remain constant if the only thing changing is the plate and screen voltage, as the output power level is changed. OTOH, if the power change is achieved by adjusting the loading on the final (with the plate and screen voltages remaining constant), the impedance will change as the power output level is varied. In some cases, the grid drive may be varied proportionally with the plate voltage as well, since for example, it would take less grid driving power to fully drive a class-C final running at 1000 volts @ 250 mA than to drive that same final at 2000 volts @ 500 mA. Hence, the Gates BC-1T/G/H gimmick of using a tertiary winding on the modulation transformer to modulate the driver stage along with the final. This improves modulation linearity, so that changes in plate current more closely follow changes in plate voltage, and consequently, changes in rf output power more closely follow changes in DC input power to the plate(s) of the final.