Ed - your Puget cite is benchmarking time to process image files. That's entirely different than the time to load a single small file from a drive, and is nearly all cpu time, which is why there is little tangible difference between the drive types in their survey.
When I talk about small files, I mean 1-2k files, or in the case of database transactions, potentially down to 512 bytes. When you do manipulations in the image catalog, there are lots of such actions. Or when you simply click around the library in LR, there is no delay from the read activity.
the time to load a file for a hard drive is largely comprised of the time it takes the head to move to the right track (1), and for the platter to rotate around to the right spot(2), and then the time to actually read the data off (3). For a small file, most of this latency is the first two, and it's why higher rpm drives perform better - the faster you spin, the less the cost of #2. That can reduce seek time from well over 10 ms to less than 5. It's also rather obvious how limiting this can be, and how valuable ram caching can be, as you repeat this cost for every load. A drive maxes out at 10s of these reads per second (50-70 being typical).
SSDs don't have heads or spinning media, so there is nearly zero latency before you get to #3, the actual transfer. "seek times" are less than 1/10 of a ms. And the transfer rate is limited more by the interface, with all recent Sata3 models doing 3x the sequential speed of the fastest drives. When we're talking about 512byte reads, they can do tens of thousands per second, rather than tens per second.
