Its the transient response of the card during voltage shifts. If you want to know more about transient response and how it affects electronic circuits I recommend watching some of Buildzoid's videos on Youtube.
The really oversimplified version is that changes in state can lead to overshoot and you can't see this in monitoring software, only on an oscilloscope. The voltage spikes you can't see can be much higher than the safe limits, eg. the card tries to ramp up to 1.081V, but overshoot could say lead it to run at 1.3V for a brief moment. It leads to crashes at least, and at worse will just burn out components.
The power delivery of the FTW3 cards can't respond fast enough to regulate the voltage changes safely - my guess is they didn't quite account for something correctly with the 3x8 power inputs or the programming for the VRMs isn't right. Best case they can reprogram the firmware and a BIOS update can fix this, worse case the card might need a redesign of its power delivery/filtering. We already know that Nvidia pushed Ampere to the limits so much that early cards were crashing, with some cheaper cards being more prone to crashing.
If you're wondering why someone would be interested enough to go listen to Buildzoid's technical ramblings, it taught me a lot about LLC settings and overclocking my CPU. I can instantly crash my otherwise stable system by forcing an extreme state transition - eg. launching Intel XTU's benchmark test. I was able to tweak LLC on my Z370 board enough to deal with a lot of instability but its VRMs were never designed for the power hungry i9-9900k I upgraded to. What you're doing in LOL is going from low load state to max speed/voltage so its a big jump in the curve - you can mask the problem by limiting the overshoot say setting 1.05v max, so your overshoot is lowered enough that it doesn't crash your system, but it doesn't solve the bad transient response on the card itself.