Now this is interesting. This is taken through a rotating mirror.

 

I joined a first surface laser mirror to one of my motors. Running at 2250 RPM and with the spark 16 cm away the radial velocity of the spark is 37 m/s. With the image being only the negative 2 cm of a total 7 cm spark width, the vertical distance of the photo is 500 us. So you should see events in the region of 10 us easily enough. There doesn’t seem to be any structure at that level around the discontinuity.

 The Tesla coil above is my junk coil running on half of a 12 kV 30 mA NST. It has a few ceramic caps and a 3 segment static gap. Primary is 15 turns and secondary is 260 turns in 11 inches. There is usually no toroid but I used one to intensify the sparks by putting an old tin on top. The gap is only about 2 cm to all fit in the mirror view. You can just see the spark in the mirror in this photo.
Of course, with each spark lasting microseconds or less it becomes harder to catch a spark in the mirror. Even with 2 second exposures and the spark firing at perhaps 20 Hz you only get a spark in view occasionally. It should be easy to increase the resolution by a factor of 10 – 20 to see events at microsecond level. It may take many minutes of exposure to get a spark though.
This would be of great interest to Tesla coiling to get sparks seen on that time frame.
Note that this is not a true high speed photograph. Vertical movement of the spark on the image may be due to irregularity of the spark or due to events happening in time. Multiple spark channels should show up well or stepped leaders perhaps. 

The left photo shows a single spark and the centre photo shows multiple sparks captured with a longer exposure. Mirror to spark distance is 38 cm which means that the image moves at about 100 m/s. The picture represents about 2 cm width and 4 cm height i.e. vertical scale is 40 us. (just over 100 ns/pixel). The right photo shows a view of a LED being flashed at 100 kHz hence the distance between each LED is10 us.
So what do we see and how to interpret it?
There is a ladder of sparks with each spark being fairly discrete and without any obvious parallel sparks. All sparks seem complete and there are no discontinuities. Almost all sparks are bright at the ends but less bright in the centre third. This also corresponds with what you see when it is running. I am not sure what it means, however, if each spark is a single cycle then the negative one third may brighter each half cycle, leaving the centre dim.

The left photo shows a Royer ZVS circuit firing a rewound inverter MOT transformer to give perhaps 2 kV at 15 kHz. It wasn’t bright enough to show so I later added a diode, resistor and .06uF mica cap to give a brighter spark which was rather irregular due to the low firing voltage. The right photo is 100 vertical pixels = 10us showing 3 sparks of less then 1us duration, which appear to deviate from a vertical line. Going back to the setup photo, you can see that one of the electrodes is vibrating changing the spark position.
The 3 sparks suggest that there is a resonance at about 10 us period – 100 kHz due to the .06 uF cap and the effective series inductance of the cap itself plus the two 8 inch crocodile clip leads. As you can judge by the pixilation (automatically smoothed by the software) plus the noise, the camera is being pushed to the limit. Very small sparks still seem to be point events. Hopefully a 2 foot TC spark will have more structure.
To see speed of light events I would need to have 500 foot events which would be 1us. In fact it would not be too hard to bounce a laser over a path this length to show the speed of light. Hmmm… I have a corner cube prism and two eight inch parallel first surface mirrors. Add a beam splitter or two, line it all up and go. Ohh, and it needs to have picosecond switching. Did I mention that? Maybe my scanner Hex mirror assembly could rotate the laser beam to give fast enough effective switching. Head is starting to hurt here.