Of course, Benching and real world gaming are affecting conversely. Benchmarks usually tells the overall numbers of a component as a score, so we can compare which is product is faster against whom. On the other hand, real world gaming performance has totally different scenario, there are different factors affecting the performance such as you mentioned "The bad port" which is developers error. For other issues, it mostly depends on whole system. Take an example of Valley benchmark, when you overclock a GPU to a level it can easily passes the test but when you try same clocks while gaming such overclocking might crashes, hence a failed overclocking cuz the gaming takes the load of whole system and not just GPU so either factor could imbalance the equation.these scores don't mean a thing if the game is just another console port.. optimized badly for the pc gamers.
wild lands is a big example it has a bad gfx world ... it has noob mesh and v bad quality overall but its hard to max out for 1080p on gtx 1070 , yet witcher 3 was better looking in everyway u can even max on 1440p.
What Valley do is to give you a number allows you to compare with another GPU, synthetically. Similarly, What furmark produces on thermal is gaming would never. Synthetic benchmarks demonstrate which is better considering the whole system ( In 3DMARK case). Beside 3DMARK and Valley, Reviewer also test many games on GPU while they are reviewing a Graphics Card. So in short, These synthetic benchmarks and numbers are just for estimating GPU worth not the actual performance and worth. Two 1070, will perform differently. Same goes to CPU.
3DMark and Valley are 20% of testing completed, rest is Gaming.