The important part of the described attack is its ability to hop past the fried switch, possibly more than one level, to affect devices elsewhere on the network, possibly hundreds of meters away. That makes it distinct from traditional ethernet killer or hammer attacks.
With about 15 minutes of research and looking at electrical diagrams and discussion with a colleague, I figured out exactly what device he's using. If I can figure it out, so can anybody. Out of respect for the author, I won't disclose it either, but I'm sure most of the Slashdot crowd could figure it out as well. The device in question is not expensive and is portable as he describes and has the right electrical properties to not fry the voltage shielding on the ethernet cables while being able to bridge circuit gaps in a sustained manner, as he demonstrates with the 4-5cm spark distance. It is also distinct from lightning strikes because of the variable duration of application and precision with which it can be controlled, which can result in more damage than a large burst of lightning.
With some tweaking, it is conceivable that a single ethernet port in an unattended area like a hotel lobby or university public area (both of which are common) could be targeted such that in just a couple of seconds, damage could be done to devices throughout the building, even devices not directly connected to the switch to which that ethernet port is wired. It's unclear how many hops are theoretically possible, but I suspect at least 2. Research in a controlled lab environment would be worth exploring.
That's a threat worth serious consideration. None of the network architecture in the many different places I have worked was ever designed with this sort of attack in mind; a fried switch was considered the worst possible scenario. This is much worse. At the very least, it should remind people that unprotected ethernet ports can be a huge risk and encourage them to improve physical security design.