Not sure how much vFlash improved with vSphere 6, but with 5.5 the function is limited.
It is not a dynamic resource. You can’t just put 2 TB SSDs in the server and say use them for read cache.
You have to dedicate some space to each VM which makes it rather... well not complex... but time consuming.
Also what happens if you move that VM to a different server and on that servers is not enough vFlash capacity left for this VM?
I also have to agree with Travis!
A good designed network (10GBit) should add well below 20 microseconds of latency.
I don't think there will be a notable difference whether the IO comes from local SSD or a Nimble SSD.