See how it's angled in during the onset of the plateau?
Suppose Cdg were massively smaller at that point; what would happen? Well, Vds would drop a whole lot faster, and Vgs would rise faster, too. Though not quite as fast as below threshold.
Now reflect upon my waveform:
Well, let's see. The initial slope is about 3:1, or 6V per 5.2nC, or 0.87nF (which is Ciss, close to the 1nF datasheet figure).
When Vds starts to drop, it gets more shallow, about 1:1 slope, or 2V/5nC, or 2.5nF equivalent. We could calculate the voltage gain and Miller effect, but better to use the totals: delta Vgs = 1.2V (4.0 to 5.2V) delta T = 3.3us --> delta Q = 3.45nC delta Vds = 340V
So delta Qg = 1.044nC (still assuming 0.87nF), so delta Qgd = 1.4nC. 1.4nC / 340V = 4.11pF average.
Which is pretty astonishingly low!
It's not fully "on" until 20us after trigger -- even though Vds swings 90% in the first 7us. That's the magic of these SuperJunction parts. The capacitance acts like a snubber network built into the MOSFET, massively saving on switching losses, even for hard switched loads!
Tim