The plasma torch needs to fire and start an arc before moving or a detail can be missed or a drop out doesn't drop out. That was the case that I ran into. Thinking that it was a GRBL issue, I posted an issue on the github grbl page here. The thought was the G4P1.2 was not dwelling the proper about of time the first time it was called, but would work fine in successive calls. The program below was made to help with diagnosing the problem.

G21 G90 G64
G54
G0X0Y0
G0 Z1.0
G38.2Z-300F1000
G92Z-17
G90G0Z1
M3S10000
G4P1.2
G1 Z1.0 F1000.0
G1 F1200.0 X10 Y0
M5
G0 Z10.0
G0 X0 Y10
G1F1000.0 Z1.0
G38.2Z-300F1000
G92Z-17
G90G0Z1
M3S10000
G4P1.2
G1 Z1.0 F1000.0
G1 F1200.0 X10 Y10
M5
G0 Z10.0
G0 X0 Y20
G1F1000.0 Z1.0
G38.2Z-300F1000
G92Z-17
G90G0Z1
M3S10000
G4P1.2
G1 Z1.0 F1000.0
G1 F1200.0 X10 Y20
M5
G0 Z10.0
M30

This code produced the following cuts:

The two left columns were ran once, while the right column was ran twice in a row. It showed the same cuts the first time through, but worked the second. So it might not be a grbl issue. So a phone was setup to capture the sequence of events as the program ran. The lights on the Plasma cutter are shown below:

Its upside down as the video shows it. Pretty straight forward what they stand for. The unit was on so the AC light was on. The gas represents when the air kicks in, the DC light, when there is juice to the plasma torch and the over temp is just that. This is located on the top of the plasma unit in the video. The relay that kicks on the plasma cutter is inside the plasma unit (lower left) and has a red light turns on.

For the first cut, the relay kicks in to turn on the plasma torch, but there is hesitation before the DC light turns on along with the plasma torch fire. After the relay shuts off, the DC light stays on for a short time.

The way that would be nice is if M66 was implemented in grbl (may have to dive into this, if its priority goes up). For now a cut elsewhere on the material will charge the DC side and allow the next cut to come out ok.