Thief has never been kind to AMD processors (without Mantle enabled) so it isn't that surprising for the Pentium G3258 to beat the Athlon X4 860K by an 18% margin when both are overclocked to 4.4GHz.
With the GTX 960, installed the performance margin favoring the Pentium is extended, as it now leads the Athlon by a 25% margin.
The overclocked G3258 averaged 55.1fps in Watch_Dogs using the R9 285, making it just 2% or 1.3fps faster than the X4 860K.
When testing Watch_Dogs with the GTX 960, the Pentium was 4% faster than the Athlon once both were overclocked to 4.4GHz.
Back when we benchmarked Dying Light in January we found that the Pentium G3258 was considerably slower than the AMD FX and Intel Core i5 processors, though it's worth noting that we were testing with the GTX 980. Unfortunately we didn't test any Athlon X4 chips or APUs at the time, but the FX-4100 wasn't much faster than the G3220.
As you can see, the Pentium G3258 is much faster than the Athlon X4 860K, so much so that the standard 3.2GHz clock speed was enough to crush the overclocked X4 860K. In the end, the overclocked G3258 was 30% faster than the overclocked X4 860K when using the R9 285.
Moving to the GTX 960 helped the X4 860K, though the G3258 was still 18% faster when comparing the 4.4GHz overclock configurations.
The G3258 maintained a solid lead over the X4 860K when testing Hitman Absolution using the Radeon R9 285. The overclocked G3258 was 5% faster than the X4 860K as it enjoyed a 2fps advantage.
Using the GTX 960 allowed the Athlon to beat/match the Pentium by a single frame once both CPUs were overclocked.