Yes, and for people thinking about buying Kepler 2GB vs GCN 3GB, they were pointed out Fermi's 1.25/1.5GB limitations as being historically bad (compared to 2GB Terascale). You're right in the sense that historical precedence doesn't guarantee the future. But you'd be stupid to not consider the correlation. The only way this would not be the case is if game developers just stopped using higher quality textures... and that seems highly unlikely with Xbox Scorpio coming out next year setting a new standard bar.
So basically you're just agreeing with what I originally posted, i.e. that 4GB may very well end up being a significant advantage compared to 3GB, but that there is no way to say so for certain.
Maybe when the RX 470 was released, it was 10+% behind the GTX 1060 3GB. But recent benchmarks indicate the RX 470 is 7.2% behind the GTX 1060 3GB at 1080p:
http://www.hardware.fr/articles/949-28/recapitulatif-performances.html
That average differential isn't enough to lower graphics fidelity at playable settings, except maybe on GameWorks titles.
You shouldn't use averages across multiple games to compare bottlenecks (VRAM and others), since bottlenecks will almost always only have a significant effect in a smaller fraction of the total number of games and thus have a limited effect on the average difference.
For instance take the 2GB 960 vs the 4GB 960. The 2GB 960 is generally agreed to be one of the most VRAM limited cards in newer times, but when you
compare it to it's 4GB sibling, it is only in a fraction of games that the VRAM limit is actually a problem (Techspot only saw a significant difference in 2 out of 11 tested games), and the average difference at 1080P is only 2.8% (and only 2.4% when looking at the 99th percentile). So based on the average across all games we would be lead to believe that there is no significant difference between 2GB and 4GB VRAM.
As for the average difference in individual game not being enough to lower graphics fidelity between the 470 4GB and the 1060 3GB, outside of GameWorks titles, I would beg to differ. In
TPU's benchmark suite the 470 4GB falls behind by 20% or more in 9 out of the 21 games they tested, and of these 9 titles only 3 are what I would consider GameWorks titles (AC: Syndicate, Fallout 4 and the Witcher 3, although the Witcher 3 was tested without hairworks enabled). Of course those 3 titles also happen to some of the most popular titles released in recent times, so I don't see why they shouldn't count.
In fact I wouldn't be surprised if going forward, there will be more AAA gameworks titles where the 470 falls significantly behind, than there will be AAA VRAM limited titles where the 1060 falls significantly behind.
I can tell you TODAY that the 4GB card will ALWAYS be able to enable higher texture settings than the 3GB card, we dont need a crystal ball for that.
And I can tell you TODAY that you are wrong. There are plenty of games out there where the 4GB card cannot enable higher texture settings than the 3GB card, since the3GB and 4GB cards are equally capable of maxing out the texture setting in said games. Sure in theory a 4GB card can handle higher resolution textures than a 3GB card, but you can never enable higher texture settings than what the developers give you (excluding modding), and since 3GB is currently fine for what developers give us in the vast majority of games, this is a moot point.