It sounds like they simply ran the games at high settings, rather than tailor the settings to try and workaround the card's VRAM limitations. Doesn't seem too strange to me.
No, I think they were doing something weird there, manipulating game or system settings in a way that either made the game think it had access to more VRAM than it really did, or prevented the use of system RAM as an overflow buffer. Just look at the results in this GTX 1060 roundup from a month prior, from the same reviewer...
https://www.tomshardware.com/reviews/nvidia-geforce-gtx-1060-graphics-card-roundup,4724-2.html
In the game Hitman at "ultra" detail settings, you see the 1060 3GB average 68fps with a minimum of 41 at 1080p, and even at 1440p it averaged 48fps with a minimum of 31. Compared to the various factory overclocked 6GB models, it performed just 9-12% behind at 1080p, and 19-21% behind at 1440p in that game.
Then, several weeks later, we see the 1060 3GB only averaging 16fps with a minimum of 6 at 1080p, and 15fps with a minimum of 7 at 1440p, paired with a vague note about "overriding memory safeguards". It was at that point performing around 80% behind the 6GB version in that game. Clearly there was some significant change to the test setup there. I seem to recall looking into that at the time and finding that they did change some setting to behave in an abnormal way, not like what one would normally run into.
Now, I could see the limited VRAM potentially being more of an issue in some of the most demanding games on a system with just 8GB of RAM or less. If the game needs to shuffle some data out of VRAM in that case, the system is more likely to access virtual memory, which could cripple performance.