You seem to be stuck on, or have missed the underlying point, of this:
amplifier configurations don't actually make much difference to the relevant parameters, and have many more downsides (like poor isolation and instability).
Perhaps you found it too brief to be worth consideration, an introductory or throw-away statement; perhaps you lack the understanding to appreciate the fact. I don't know. (And you aren't telling us what you lack to make the connection, leaving us to guess at best.)
A current amplifier + transformer has more downsides than a voltage amplifier -- which
may still have current gain anyway, and which may not require a transformer at all.
CE amplifiers have more freedom of compensation/neutralization, feedback / distortion reduction / gain reduction, and impedance matching, so we tend to use them most often.
Sure, let's talk h_FE. A BJT might have h_FE of 100 at DC. It might also have h_fe of 100 at say 10kHz, but it's dropping by 1MHz (-3dB or so, maybe?) and runs out at f_T (h_fe = 1), say 100MHz. It's a dominant pole cutoff (more or less due to C_cb * R_bb' predominantly I think, but also physical limits), and independent of circuit limitations which may cut lower (like C_cb * R_L setting output GBW). An emitter follower is a terrible (general purpose) choice at 10kHz already because who even makes (let alone budgets for) transformers at that frequency, but it's even worse at 10MHz where gain is dropping sharply. (Adjust numbers appropriately for any given part in question.)
Riddle: how do you get power gain from a transistor above fT? By your premise, this is impossible! In fact it's not (but it isn't terribly useful, the efficiency or bandwidth is poor).
You imagine that the difference might be distortion, or noise; it is not, the device has the same fundamental input-referred noise limits regardless of topology. You get say 1nV/rtHz of noise in series with the base, and it doesn't make the tiniest bit of difference whether the base-emitter is wired to the left or right side of the input, it's always and necessarily in series regardless!
Put still another way: the noise figure of a device
is fundamental to the device, regardless of arrangement or topology, depending only on the bias and impedance matching (and frequency range, since noise generally doesn't have flat frequency statistics). You put in a low-noise transistor, you bias it up as needed, match it to the source and load, and
that's what you get. Sure you might get lower nV/rtHz if you run it hotter (higher Iq), but you get higher pA/rtHz (current noise) too: it just means a lower source impedance for noise match. Meanwhile, the noise figure hasn't changed much.
RF design is all about eliminating confounding specifics, like the voltage or current in a circuit, and considering the flow of power back and forth through a network. In general, we don't care about impedance, because impedance can be matched with simple RLC networks, transformers at the worst -- and if impedance isn't matched to its ideal value (give or take acceptable tolerance),
it ain't our components to blame! This is why we have dimensionless figures like gain and noise factor, s-parameters and etc.
Tim