Hi Robert,
the way an oxy acetylene cutting torch works is the small outer circle of "spikey" flames heats the metal until the surface is just starting to melt or very close to it (something like that). Then the big hole in the middle of the nozzle is the oxygen hole and when the oxy starts coming out the metal is literally on fire. It's nothing more than heat and a continuous supply of oxygen that causes the metal to rapidly burn (not melt) away. So likewise air has roughly 20% oxygen and you get the same effect when the plasma torch is turned off but to a much lesser degree and of course it's not enough to keep itself going so the metal stays "alight" for just a short time after the torch power is removed. Bit of a sidenote but when I was in industry we were always told about how oil could spontaneously ignite in the presence of pure oxygen. I could never figure that one out because the triangle of fire requires fuel, oxygen, heat (where's the heat). But that's what we were always told.
My suggestion of a smaller relay was more for consistency of time to de-energise. For example if say a big dopey relay and a small fast relay had variation times of 20% then 20% of 3mS is a lot less than 20% of 15 mS. Funny thing about the relays, I just bought two 5v reed relays yesterday. When I get around to it I'm going to hook them up and test the deactivation time. They should be even faster than the PCB relay I use on my hole centre marking PCB. If you use one make sure both the current AND power handling capability of the contacts is more than enough. You also want longevity so the relay doesn't want to be stressed to its' max.
Don't know if you haunt Plasmaspider, but the divot issue has came up over there numerous times and a few have got rid of them with synchronously switching the torch off a short distance before the end of the cut path. The good thing about UCcnc with it's excellent M10/11 (laser output) and excellent M205/206 (THC ON/OFF) you can even disable THC just before you give an M11 to turn of the torch so loss of ARC OK will not halt motion in the last mm or two.
Also I tested the repaired version of the UCcnc install yesterday and they've fixed the glitch in the step pulse stream where you put an M205/6 and a change in feedrate at the same place in gcode.
Balazs also got back to me and said they are going to implement the changes I suggested so you can adjust THC Correction speed and Corner Anti-Dive settings while cutting. That is the ultimate method for fine tuning things in my eyes.
Keith