The motivating factor is usually the junction capacitance and operating (V, I) range of the finals.
Back in the toob days, you could choose anything from 100V (TV sweep tubes) to several kV (proper "transmitter" tubes), and up from there at higher power levels. (The largest single linear amplifying device in history* is a ceramic transmitter triode, rated for around 30kV and 30A, capable of 2MW signal output in class C operation.)
(*Hmm, correct me if I'm wrong on this. MOSFET dies only go to a few kW in linear operation, and the rest are switching devices: IGBTs, SCRs and so on. Most of those are multiple dies in an integrated module, too, not necessarily "single". I don't think cold-cathode field emission and ionization devices can do linear operation, but I've heard really very little about them. There may be larger magnetic amplifiers; in that case, I might add the condition: largest in watts * maximum operating frequency.)
Anyways, tubes have the predominant limitation that their transconductance and load conductance are very low, relative to capacitance, and to other more fundamental limitations of the device.
The bandwidth limit is given by the load resistance times the plate (or collector or drain) capacitance.
More bandwidth is desirable, because the transmitter needs less precise tuning for a given operating band, or can span multiple bands.
It turns out that the best transistors -- in terms of bandwidth, efficiency, power capability, and generally being reasonable to operate (not requiring dangerously high voltages or annoyingly high currents), fall in the range of 30-200V (and as many amps as the junction width and power dissipation allow for). For higher voltage ratings, the junction is physically longer (i.e., the path the current takes from "+" to "-" -- MOSFET channel length or BJT collector depletion region). Larger distances mean lower frequency limits.
There are some high voltage RF MOSFETs available, but they're usually rated for only 30-100MHz, for Vds in the 1200V to 600V range. (Whereas a regular LDMOS transistor rated for 120V and 10A will do low GHz without a problem.)
I'm not aware that anyone is making power PHEMTs or MESFETs or GaAs/GaN FETs in high voltage ratings, for microwave service; there are GaN power switching transistors, which are relatively new, that are much faster than Si MOSFETs for the same applications.
So, where your question fits into all of this:
- For a given device, operating it at lower voltage, and the same maximum current limit, requires a slightly lower load impedance. You have less V and same I, so V*I = P is lower. The reduced load impedance can extend bandwidth the same amount (but whether that is available as signal bandwidth, depends on the design of the rest of the circuit).
- For an optimal device, you can match to a lower load impedance, and draw more current, keeping power constant.
Too-low impedance is also undesirable, due to stray inductance having the same bandwidth-limiting effect that capacitance has at high impedances.
Tim