1. I log into service X with account A1 via Mullvad from country C1.
2. I log into service X with account A2 via Mullvad from country C2.
If the service wanted they can calculate how likely it is that A1 and A2 are the same WireGuard key. If you only use one exit server this probability won't be very precise. But the more exits you use the more accurate it will be even if the sets of exits are distinct between the two accounts.
If the egress IPs were assigned randomly all that service X would know is that these were both Mullvad users but the IPs alone wouldn't allow them to correlate the two users further than that.
I have been confused by this mitigation because switching networks while using the same service is pretty much always a VPN. But maybe I'm not aware of another case where that would happen?