The problem is most sites don't know how to test.
1. Every system is limited by whatever component bottlenecks it. This will usually be the GFX card.... add a 2nd card in SLI / CF and things change, the GFX card if oft no longer the limiting factor.
2. The impact of RAM is far greater on minimum fps than it is on average fps. Again, most sites don't bother to test minimum fps.
3. Like the site you referenced, they focus on price which is not static. Most mindsets are based on early DDR3 testing when yields on high speed RAM were low .... so whether it's worth it to spend an extra $120 for 16GB of 2400 is a different question 3 years ago, than whether it is worth it to spend $14 extra today.
4. That same argument is also usually bolstered by false logic ..... it's not worth it to spend 100% more ($120 => $240) on RAM, for fps increases averaging 0 - 11%. Your RAM doesn't go faster, your whole system does. So 3 years ago, that argument for a $1400 system should have been "is it worth it to spend 8.5 % more ($1520 / $1400) for that performance increase. Today the question is "is it worth 1% more ($1414 / $1400) cash for that same % performance increase ?". That is what I call the proverbial "no brainer"..... To put in perspective, 11% is about the same % increase as moving from a 970 to 980 and that costs a lot more than $14 ($230).
5. The difference with RAM versus GFX card upgrades is that most games will be affected more by the card than RAM as most games are GFX not RAM limited. Remember, my statement had a qualifier .... I said "even in gaming" ..... of course if you do anything else with your box (CAD, video editing, etc),the faster RAM is an automatic..... if used in a production environment or if your time has any value, it will pay for itself in no time. Metro 2033 gets a miniscule performance increase going from 1600 to 2400 (116.8 / 116.7) ~ 0.1% ... on the other hand, F1 gets 11% (177 / 159) ~ 11.3%