That's what sites like GameGPU do. How else did we find out that 2GB wasn't enough for many titles? It's because 2GB cards tanked while 3GB cards didn't. When there is a VRAM bottleneck, the performance drops so dramatically or the game experiences constant stutters that it becomes immediately know it's a VRAM bottleneck, not a memory bandwidth or GPU bottleneck. You should know what a VRAM bottleneck is like since you have 680 2GB SLI (i.e., you have a lot of GPU horsepower to turn up settings/textures high enough to run into massive VRAM stutters/fps drops. Once you do, you'll never forget what VRAM bottlenecks feel like because a game becomes a near slide-show or stutters so bad it's 100% unplayable). For a professional reviewer, it wouldn't be hard at all to compare GTX970, GTX980, R9 290/X and see if there are big stutters with the 970. Since 290/290X and 970 are pretty similar in performance and since 980 has the full 4GB of VRAM, it would be extremely easy to figure out which game
needs more than 3.5GB of VRAM from direct testing.
So far, none of the sites that noted 2GB VRAM bottlenecks have let us down. With Digital Foundry, GameGPU, TPU, Sweclockers, Guru3D, TechSpot, there are so many sites that will provide sufficient information to show when 3.5GB will become an issue vs. 4GB and the same on 4GB (Fury X) vs. 6GB (980Ti).
BTW, you do not need necessarily have to test 2 identical GPUs to diagnose a VRAM bottleneck. Sometimes it's just obvious because the performance falls off a cliff (mortal komba t x) or performance is a practical slide-show such as in Batman AK or F1 2015 on a 2GB card:
http://forums.anandtech.com/showpost.php?p=37892195&postcount=12