[jan-jaap]

This might be a dumb question but, could the loopback test failure simply be a symptom that the ethernet controller is generally unable to function?

For example like having a corrupt MAC address

This is actually not dumb at all.

The MAC address is stored in a 53c556 serial EEPROM on a little board with the reset switch (SGI made a habit of linking the system serial number to the chassis, not the mainboard). It is connected to the E-module with a ~ 10 wire flatcable.

The photo is from a 4D/25 or 4D/35 but I suspect they're all the same.

There's an internal and an external ethernet loopback test. I assume the external loopback test requires some sort of loopback cable attached to the network port, but the internal loopback test basically sets up a packet or frame for transmission, and checks the RX buffers for the size and checksum. This will almost certainly require a valid MAC address.