Chuckey and Mtwieg: you're both right!. I don't mean to be obstinate on my reasoning, I don't like to walk opposite to the rest of the world, just need to convince myself of the cons on my approach. It is: my freedom to make the transistors see an adequate load impedance is adjusting the number of turns on the work coil. What could be simpler?. It worked fine in a 50kHz 1kW prototype, no cooling necessary. Yes, N was something messy (29 turns in 3 layers, wire didn't self-support) but not impractical.
Could it be that water cooling becomes mandatory at higher powers? (pipes with small OD wouldn't exist). Could it be the additional series inductor helps comply with harmonic radiation regulations?. Maybe I'd rather read a book on induction heating basics.
Thanks all of you, and have a bright new year!.
P.D. Yes, there IS a danger on the simple LC: series resistance at resonance is very low without a workpiece, but I just shift the driver's frequency higher when this situation is detected, and no transistors blew.