I didn’t trust that the speedtest and Edge weren’t adding some hidden overhead, so I just did a quick test with iperf. I also didn’t trust that disk accesses (e.g., from paging) weren’t contributing to the host load, so I doubled the VM’s memory from 6 GB to 12 GB in this test.
The VM acted as the iperf server. Nothing else was running except task mgr.
Running a 30-second test 3 times yielded an average of 913 Mbps. Host load varied between about 305% and 345% when a test was in progress. VM’s task mgr reported between 20% and 25% utilization (normalize to 80% to 100%).
So, lower throughput but also lower load versus the speed test.