Today, finally I was able to set up stable Gigabit Ethernet connection to my laptop. Did some quick test by sending random file from my laptop's SSD to CM4 and storing in /dev/null. Here is the result measured by nload:

And I got quite nice ~700 Mbps. I still hope this is not maximum for CM4, but SSD was bottleneck this time. Let's see what iperf is gonna say. And it showed this:

By the way you can see dd counting previous attempt. So, here we have ~860 Mbps. This I can believe in.

Then the summary is that my board does its job nicely. I played quite a lot with trace length matching tool in KiCAD to make sure that all traces carrying Ethernet signals are of exactly same length. I'm curious what would be the result if I ignored that. But I think that spending money and time on next batch of test boards is not worth it :)