That is probably the nic in said machine still being alive and unable to pass any received frames up to the os. When its buffer runs full, it starts spamming pause frames, which dumber switches just blast out every port effectively paralyzing the whole network segment.
You could verify this by running Wireshark on another machine, or try disable sending pause frames on the machine before it crashes via eg ethtool
In my case just a few weeks ago I had an AP that went bananas, and was directly connected via ethernet to a RPi4, and made the Raspbian kernel panic on the Broadcom driver, randomly. Not long after the AP stopped working completely.
You could verify this by running Wireshark on another machine, or try disable sending pause frames on the machine before it crashes via eg ethtool