The only thing you did not compare is the overclock-ability. The Maxwell does not overclock with CUDA applications and as a matter of fact with any CUDA application it down-clocks the Memory speed. The new Pascal cards can readily have the Memory Clock boosted and while it does down-clock slightly when you run any CUDA app it really can boost performance. I do not have access to either Titan X but with a simple GTX 1060 SC (EVGA 06G-P4-6163 KR) I can boost the Memory Clock which in GPU-Z says on the Graphics Card Tab that it is 2002 MHz (but with a CUDA app down-clocks to 1900 MHz). With overclocking on the Graphics Card Tab of GPU-Z I can easily set my Pascal card to 2500 MHz which when you look at the Sensors Tab shows that actually under full CUDA load it is running very conservatively at 2400 MHz (or as nVidia would specify it at 9.6 Gbps). You can see below that under full load total Power Consumption is only 51.5%.
The two bursts of full utilization are the two GPU accelerated tests in my Premiere Pro BenchMark (PPBM)
I replaced my GTX 980 with a Titan X (Pascal) and all my CUDA accelerated timelines exported approx twice as fast. One timeline with 4K exported to 1080 and one timeline 1080 with Lumetri Color, exported to 1080. All exports were done to the CineForm codec.
So it can really make a difference with the Titan X Pascal card.
HP Z440, Intel Xeon E5-1650v3, 64GB RAM, SSD drives.