Hello @bergipau
I'm a little surprised you're managing to get a "great" mesh connection rating with concrete obstructing the paths. In any case, it's important to understand that there's a single 5GHz radio in these units. That radio is used to carry traffic between them via a (hidden) mesh network. But, that same radio is also used to talk directly to 5GHz WiFi client devices. So, any traffic from those devices that are connected to a secondary (i.e., either your Nest WiFi Point or your Google WiFi unit) will be going through that radio twice. That means it's going over the same channel twice. If that client is close to one of the mesh secondaries, that data may be transmitted at a very high rate. But, then the mesh unit has to retransmit it back over the same channel at a lower rate that can get through the concrete to the primary/router unit. We can model the expected end-to-end throughput as something like 1/end-to-end = 1/client-to-secondary + 1/secondary-to-primary. Plugging in some typical numbers, if client-to-secondary is around 400Mbps and secondary-to-primary is around 120Mbps, then we would estimate the end-to-end would be around 92Mbps, which sounds like about what you're seeing.
You can get more details by going to the Google Home app and finding one of the devices you're testing with, opening the details screen for it, and switching to the "info" tab. First, verify that device is actually connected to the access point you expect (that's another complex discussion). Then scroll down and take a look at the three numbers under "Device Speed". The top number in that list is how fast your primary/router unit can talk to the internet. The middle number in that list is how fast that mesh/secondary unit was able to talk to the primary during the most recent mesh test. The bottom unit may not be filled in until you run a device speed test (the "Test device speed" button just below that), but it is how fast that device can talk directly to the access point it is currently connected to.
Interpreting the middle number can be tricky, since that test is running for all mesh/secondaries simultaneously, so it's not necessarily the speed you'd get when only one unit is carrying traffic. But, it's still a good way to compare different unit placement options against each other.
I hope some of this extra detail helps. Mesh networks like this are great for expanding useful coverage area, but they often do that by sacrificing peak performance in some areas for larger useful coverage.
