Notes on the simulation circuit:

1.  The transistors chosen are for availability in LTSpice and not physical 
    availability. Provided some basic parameters are met, IT DOESN'T MATTER 
    WHAT YOU USE.
    TR11 and TR12 need to be 100V 10A pair, matched for preference and at least 
    100kHz Ft.  (a 2N3055 does NOT work (Ft 10kHz))
    TR9 and TR10 need to be 80V 0.5A pair, matched for preference.
    TR4 and TR5  need to be 40V 100mA fast RF transistors.
    All the others need to be 80V 100mA signal transistors 
  
2.  C5 at 33pF gives a 3dB roll-off at 120kHz.  330pF has a 55kHz roll-off

3.  D5 in the original circuit was a gold-bonded germanium diode but a 
    low-foreward-drop Schottky diode appears to work and is much more available.
    It is used, along with a normal silicon diode D$ to bias the 
    signal-splitters TR4 and TR5 to near conduction to improve the transfer 
    rate across the zero-point.
    
The "thump" on startup seems impossible to remove.  One might have to put a 
    delayed relay to connect the loudspeaker after 5 secs. (must run the circuit
    with a resistor on startup to charge C7 and changeover to the loudspeaker).
    Blomley ,entions this and suggests increasing C3 but even a 1F capacitor 
    (because you can in a simulation) does not prevent it entirely.
    
R3/R5 determines the overall gain of the amplifier.  Blomley set this to 100 
    using 4k7 / 47 resistors.   
      