Well, it depends on what you mean by "implementing".

I believe we can get the code working in a few days, and there are probably people already working on a PoC of this (or one of the alternative proposals). So if this is what you mean by implementing, the answer is "days".

But as anarazel is pointing out, this is going to touch a fairly critical part of the database - it interacts with storage, MVCC and likely various other things (e.g. various index optimizations like Index Only Scans).

Moreover there are other proposals (I'm aware of 3 or 4, and I might have missed some), so the question inevitably will be - which of the proposals is the best compromise?

So I expect a lengthy discussions on pgsql-hackers, a lot of testing, benchmarking etc.

But all this does not really matter that much - there's no chance this could get into current releases, so the earliest release it can get into is 10, which means code freeze likely sometime in April/May 2017.