4 off 300Bs in Parallel Push Pull will get you close to 40 Watts. I'm about to do something like that myself.
Why - coz I have these Output Trannies which are currently being driven by 4 off triode strapped EL34 but deserve something better -
http://www.plitron.com/wp-content/uploads/P4006-01.pdfThe trick for decent parallel push pull triodes is a "GRUNTY" driver stage to drive the Miller Capacitance of the Output Tubes to high frequency without slew rate limiting.
Math:
mu 300B = 3.85
Cg1 = 17 pF
Cg1a = 7.7 pF
Capacitance at grid 1 each 300B = 17 + Miller Capacitance = 17 +(3.85 x 7.5) = 46 pF
2 in parallel + wiring strays - call it 100pF to drive on each side of the push pull
The 300B grid will bias at around -80V which means that we need to swing 80 Volts peak (160V peak to peak) signal to the grid.
Depending on which data sheet you read the Plitron -3dB roll off is either 187kHz or 217kHz.
That means we want to drive 80V peak into 100pF to a frequency of at least 200kHz (say half this for ordinary EI Output Trannies)
Slew Rate = 2 pi F x V peak / 10^6 Volts per microsecond = 2 pi x 200,000 x 80/10^6 = 100 V/us
Signal Current = Slewrate (V/us) x Capacitance (uF) = 100 x 0.0001 Amps = 10mA.
That is the signal current required so the idle current on the driver stage would want to be say twice that = 20mA, maybe a differential amp made up from triode strapped EL84 or 6V6 or similar.
This should tell you that this sort of amp is not trivial to design but taking a BFI (Brute Force and Ignorance) approach is not a bad way to tackle it.
If you don't want to get involved with slew rates etc there is another way to look at the problem:
200kHz means a signal time constant of 1/(2 pi F) = 0.8 us
Sampling theory tells us that for a signal to settle to 99.7% ( i.e. 0.3% distortion) of its input value takes 5 timeconstants, that is in this instance we want the grid drive 5 timeconstants to be 0.8 us so actual timeconstant = 0.16 us
This timeconstant will be the output impedance of the driver x the 100pF grid capacitance.
Thus Driver Output Z = 0.16 us / 100pF = 1600 Ohms (max)
Cheers,
Ian