There are three places I notice the speed differences: booting/rebooting, launching games, and transitioning between levels within a game. Boot times were significant, on the order of 12 seconds after POST to the desktop versus 20 seconds; not world-shattering but definitely noticeable and only moreso as you load down a system with drivers and apps, etc. Game launch differences were pretty similar but there's so much unpacking/decompressing/malloc'ing going on then that it masks some of that resulting in gains that may or may not be subjectively noticeable if you're busy grabbing a drink or fiddling with something else. Where I really noticed the difference was in level transitions in games like Half-Life Ep. 2 or Supreme Commander, and to a lesser extent zoning into Orgrimmar. The thing is, it tends to be cumulative and while you might not notice your load times going from a minute to 2 seconds, my experience was that the overall quality of life was well worth the upgrade.

That said, once you pass the 1GB/sec boundary, I presume the bottlenecks are now memory and interconnects -- I didn't really notice a lot of difference between two 840 Pro SATA 2.5" models in RAID 0 (usually gave me 1-1.2GB/s r/w) versus a single 950 PCI-E m.2 except in random writes where the m.2 would stall on me. Moving to two 950 Pros was very noticeable again, especially in writes. Copying a 3.4GB Linux ISO around in a few seconds just gives me a warm fuzzy.

In hindsight, it would have been much less hassle in terms of UEFI install shenanigans (and probably cheaper at the time) to have just picked up 3x 850 EVO SATA 2.5" drives and run them in RAID 0. That would have put me at ~1.5GB/s read and write versus 1.5r/2.4w on the PCI-Es, which would have been much less of a difference. I'd do some measurements for you but honestly, I am just unmotivated to go tearing apart my drives again.