1) 2009 Mac Pro is a 4,1, not 3,1 (as per your title). This make a huge difference. Because there is a power management bug in the 3,1. Make all the HD7xxx card under perform. But on a 4,1, the HD7xxx card is one of the best choice (especially your usage is FCPX).
2) I assume you are running 4,1 but not 3,1 (because of the HD4870). The "Best" GPU can run is TitanXP. However, that's "best" in general, not in your case, and not consider the cost to performance ratio, or OS requirement, etc.
3) For Photoshop, GPU is virtually irreverent. It's a CPU limiting software most of the time. All you need is a fastest CPU (I emphasis "fastest", not high core count). I can tell you that the Photoshop speed test is virtually identical with HD4870, single HD7950, dual HD7950, R9 380, or a 1080Ti. I done that myself. But upgrade the CPU from W3520 to W3690 (2.93 vs 3.73GHz, 27% faster in single thread performance) will make the speed test complete 27% faster (~11s vs ~15s).
4) For FCPX, the "best" at this moment is RX580. HOWEVER, you need 10.12.6 at least. Which is NOT directly supported on a 4,1.
5) If you want to keep it as a 4,1 (but not flash to 5,1). And only run "supported" OS, then I suggest R9 280X is the best in your case. This card is cheap, very easy to flash (make it act like a Mac Edition card, which can display boot screen), and perform extremely well in FCPX (it has exactly the same device ID as the D700, the high end GPU in the Mac Pro 6,1. That's why FCPX is highly optimised to this GPU).
6) Another "best" option is go for dual GPU, 2x HD7950 (or R9 280, NOT 280X) can beat any single card in FCPX. I've done
that before.
7) This "best" is base on rendering (and compute) performance, not preview performance. If you want real time preview performance (with no pre-rendering, no proxy, etc), then R9 280X is bad. 1080Ti can do much better job.
