You don't directly drive the output stage gate of a FET (or BJT for that matter).. Or course it is too big. There are a series of transistor stages that builds up the gain before driving the final stage. Your input stage is a small transistor with high gain and low input capacitance. You get the right gate drive levels by end-to-end feedback loop and the middle stages are sized correctly. This is very standard, this is not even my field but any design on the back of a magazine that I saw when messing with this stuff 20+ years ago had this topology.
So I'm not really sure what you are referring to. Maybe there were a few hokey products back then but let's not confuse the issue - FETs are just fine.
The amp in question that DOES differ from the normal topology using MOSFET output transistors is Acoustat TNT 200 (and subsequent models ) designed by Jim Strickland. IIRC, the driver does NOT drive the gate of the MOSFETs - it was found that this topology is much easier to drive than the standard one.
http://www.diyaudio.com/forums/solid-state/102498-hey-found-acoustat-tnt-200-schematic-tnp-preamp.html
I agree there must be some smaller transistors before the grand finale - I have yet to study the SRM1MK2 in full. As it is an amp using by now obsolete parts ( high voltage transistors used are normally used for horizontal deflection in cathode ray colour TV screens; since TVs have become pancakes, high voltage parts are ever harder to source ), I try to mess with it as little as possible. The reason why it behaves so to prefer really low output impedance "devices" supplying its signal, means there must be something frequency dependant (reflecting back? - remember, load is pure capacitance) at its input. As it is, I prefer having to drive it with a low output impedance source to doing anything else - it works OK this way for me.
Believe me, at this moment above is the least of my concerns. My Lambda Pro developed a problem ( loss of sensitivity, not that uncommon after 30 or so years ) and now I am gathering together the courage to try to open and repair it WITHOUT tearing the driver diaphragm. Having built my own electrostatic earspeakers, I am familiar with what had happened, problem is that the driver of the Lambda Pro is NOT designed to be rebuildable - basically, one is at the mercy of the glue(s) holding everything together - it can fall apart all by itself (documented on head-fi) - but it could equally tear the diaphragm . Only one way to find out ... - each dab of the glue holding the driver together was applied by hand and there is no way of telling how much force will be required to pry the driver open.
Original Lambda Pro replacement drivers have not been available for years, if not decades
The reason why I do not want any other model of Lambda than Pro is the fact that I own Diffuse Field Equalizer Monitor 1 ED - tailor made for Lambda Pro. Any change of diaphragm ("easy" to do, lighter, faster, allowing lower resonant frequency and better dynamic range ...) means its frequency response will no longer be compatible with its dedicated equalizer - something I would like to avoid at all costs. Lambda Pro is getting exceedingly rare for sale, its price is slowly but steadily going up, it is 30 or so years old by now - and that means it could happen again with a "new" set. That is why I am slowly building the courage to try to open it without damaging it - it was never meant to be done, replacement of the driver only. Subsequent models of Lambda have adressed this problem - much easier for repairs.
P.S: There is a schematics for the TNT-200 amp - on page 3 of the thread from the link above. It is driving the gates after all ... - but it is different enough from normal to allow for the patent to be granted. United States Patent US4467288