Thank you all for your help, much appreciated!
I brought down R4 to 470E, which reduced the switch-off delay to about 700 ns:
Ch 1 yellow = input on J1 -1, Ch 3 magenta = Drain Q4, Ch 4 green = bases of Q3 & Q4
I coincidentally found that disconnecting probe 4 (green) reduced the switch-off even further to about 550 ns, but which an undershoot of -12V:
Apparently the probe capacitance has a big influence!
Adding the C2 of 100 pF across R1 as in schematics of Zero999 caused by far the largest improvement; delay of just 200 ns but with an undershoot of -15V!
I don't understand where the undershoot is coming from; the inductance of the wirewound load, resistor R5?
To reduce the undershoot I added a 100 pF from the bases of Q3 & Q4 to ground, now 'just' 5V undershoot for a slightly increased delay of 350 ns.
But this is somewhat silly I think, adding a capacitor slows things down again ...
So I removed it and added a flyback diode 1N4007 across the load instead. This removed the ringing all together, but still an undershoot of -7V!
How can this be?
What is the best way to deal with this undershoot?
@ Zero999 : As the switch-on and switch-off delay of about 200 ns is now ok for my purpose, I will probably not build your more elaborate solution.
But for my understanding, is the following reasoning correct:
the Schottky diode was removed because of R1 = 100K instead of my 1K8 doesn't drive Q1 in saturation and C2 speeds up switching times considerably.
The 10K from base Q1 to ground was also removed. I always thought it is good practice to add a resistor from base to ground in order not to have a floating base (when no signal is applied), not?
The sole purpose of Q4 is to speed up switch-on time, as it bypasses R3 to provide a larger base current for Q2 to supply charge to the gate.
D1 is needed to discharge C1.
PS: I don't understand why my pictures don't show up inline as yours do, I did read the instructions but it doesn't seem to work.