It looks like it works not by constantly underextruding, but only intermittently. The parts where the strands join are overextruded if anything.

I'm pretty sure it's actually just a continuous under extrusion. It becomes periodic in nature because material gathers under the nozzle and then it hits the blob from the previous layer and gets pulled off. It's a similar effect to how droplets form and fall off of your faucet when the flow rate is low. This also leads to the diagonal angles seen as the Z-height goes up.

It's usually hard to get this to happen consistently, which they've done quite well.

I've got no citations, but I worked in FDM Additive Manufacturing for a few years, and spent 4 months of that designing extruders for a name brand company.

What surprises me is quite how consistent it is. The blobs look perfectly consistent over thousands of layers, which seems very unlikely - even a tiny perturbation would grow bigger in each successive layer and eventually cause the blobs to be chaotic in nature, even if the first row was consistent.

I wouldn't be so sure, I saw a video of variable infill extrusion and it looked somewhat like this, although over a greater distance but otoh much faster. Speaking of which, it seems to be printing quite slowly.

From my limited expirience, temperature of the extruded material must be in some narrow window, too cold or too hot and it will clump. Even within the window, the production of sticky hairs (what one would aim for in this case) varies a lot with temperature.

I can also imagine that the material plays a huge role, but I cannot say because I've only really used PLA.

The desired result is that you get a series of pulses of extrusion rather than random under-extrusion, so if you can't get consistent pulsing by adjusting temperature and extrustion rate, you could always hack Marlin to do it on purpose. I expect that is what they've done.

here's the author explaining how to achieve the print: https://twitter.com/FormanForm/status/1315759308450062336?s=...