In compute intensive games it can be a lot more than 10%. The execution resources available in a haswell core are significantly "wider" than a PD core. (approximately double). Real-time workloads run much better on the big-core intra-core parallelism level. Users who make the switch from many-core to big-core for compute intensive games consistently report significant changes in game-play smoothness etc. The only way to "hide" or shrink the difference to ~10% or less would be if the render workload were adjusted to solidly plant the bottleneck on the GPU configuration at all times at a relatively low FPS (~30). If your idea of smooth game-play is 30FPS, then yes, a switch to haswell may have very little effect. If you're like most competitive gamers, who notice FPS minimums when they drop below ~50-60FPS, then you will likely benefit from a switch to haswell by significant margins. If you work for the AMD marketing and deception department, then you would find a way to demonstrate that the FX chips perform similarly to the competition. Easy: hide the difference under the rug of a GPU bottleneck.
When the news media takes White House propaganda/talking points and turns it into news, I am ashamed of our media. When freelance forum participants take the bullet points of the hardware marketing departments and basically share them as fact without question, I am ashamed to wear the badge of "geek."
Clock for clock, haswell can indeed produce up to ~75% higher minimum FPS in many games, though in some cases, it can be even higher than this due to the way draw calls can be slacked off in some game engines to preserve a cohesive timeline. This depends on how the game engine functions are tied together to the timeline. The compute overhead available above and beyond the minimum required to keep up with the timeline of events/AI etc, can, in some game engines, winds up having a dramatic effect on FPS. I've observed this behavior in testing. Stock clocked a CPU might cause FPS minimums to dip as low as say 20FPS, but then a mild overclock of just 10-20% could improve those minimum FPS to 30-40FPS, because that 10-20% of compute overhead actually nearly doubles the available headroom for draw calls. This may seem counter-intuitive, but many games do wind up responding this way to changes in compute performance. The substantially wider haswell instruction engine offers a much larger piece of compute overhead pie for draw calls, and thus, solves those nasty FPS minimums and glitches in MANY games.