The graph seems to indicate he did 3 tests, one using 4 physical cores, one using 4 logical cores, and one doing the two previous tests at the same time in two different vms.
These are the results I'd expect as he is pegging each core at 100%, resulting in slower throughput for each VM.
Watch the video again and listen to Luke explain how logical and physical cores exist together. The logic buddy of core 0 gets to use the extra resources of core 0, but what if that core is pegged at 100%, or even if it starts to hit numbers above 70%? Well, there aren't many resources for core 0 to share with its logic buddy. Windows tries to load balance, as luke said, hence why one VM doesn't just completely lock up when its physical buddy is using most of the resources.