Ethernet seems far easier to prototype with. There's almost no off the shelf stuff for talking to RGMII whereas Ethernet you can just plug into your laptop for testing. If it's two different teams building things it seems like it would be a lot easier to just agree on Ethernet as the interface and then delay integration testing or release earlier.
RGMII requires way more work to run board-to-board (heaps of signals, quite precise length matching, impedance control, etc. on the boards, better board-to-board connectors etc.) and at the end of all that will likely be less robust than just running Ethernet. I'd much rather use SGMII just because it's far fewer signals to match (even if it runs way faster) instead of RGMII.
The chips they're using might already have Ethernet PHYs built in anyway which might also be part of the reason they're using Ethernet.
Assembly isn't free, either an engineer or the PCB fabricator has to put that together. Also the design isn't free and it's certainly not necessarily going to match the behavior of the device on the other side.
But your laptop's Ethernet adapter comes free with your laptop (both in terms of money and waiting to get it since it's already on your desk) and possibly even more importantly you know the laptop manufacturer and users have QAed it for you so it's absolutely going to behave the way you expect which is important when the device you're designing isn't behaving.
> Assembly isn't free, either an engineer or the PCB fabricator has to put that together
> your laptop's Ethernet adapter
The device as-designed likely wouldn't work with your laptop's ethernet adapter - hence why the author of TFA placed an isolation transformer and jack ...on a breakout board.
RGMII is not some obscure competitor to Ethernet, but rather, Ethernet was designed to be a modular two-part design with "MAC" and "PHY" chips connected via "MII" interface. RGMII is simply the latest version of it.
Many Ethernet-supported SoCs still use various MII style interfaces because it makes more sense to outsource the physical layer to some external chip especially if not everyone is going to use Ethernet.
It's perhaps like the difference between using Thunderbolt vs raw PCIe. You technically shouldn't need Thunderbolt if you're just permanently connecting two things inside a same machine.
Is it smarter to do it proper and make it silicon efficient than just shipping the darn thing ASAP? idk. We'll see.