Graphics APIs are really good at setting up async transfers and game engines are good about fetching stuff in advance of when it's needed. So, I think it's mainly the utilization peaks where you can find any savings, upon moving up to a faster speed. In the above link, you can see this in how the 1% minimum FPS benefits more than the mean FPS.Agree, if something takes 16 seconds then doubling speed it takes 8, doubling again.. 4 then 2 then 1..
The returns yield diminishing time improvements each time. Hardware always lags the connection specs.
As for reducing command latency, you're right about diminishing gains from doubling speeds. Commands aren't responsible for the bulk of bus utilization, though.