Why was move 16 not a blunder/inaccuracy?

So, analysing my latest game, I'm struggling to see how move 16 wasn't at the very minimum an inaccuracy?

http://en.lichess.org/PK7T88Dz/white#39

Qd4xa7 did win me the game a few turns later after my opponent made several unforced mistakes, but Qd4xRh8 objectively seems the much stronger move, putting me a rook/knight up and stopping him pairing up his rooks, as opposed to knight/pawn up, with no forced mate immediately apparent.

I did not look at this in much detail. If you play Qd4xRh8 the queen is trapped after Nf6. You would get both rooks for the queen which is in general a good trade but in this position the initiative with the queen seems stronger in the engines view.
You can use the local computer evaluation button over the notation to check out some lines.

16. QxRh8 loses your Q after 16... Nf6
Your Q is trapped and threatened by Rxh8

I was aware of Nf6 but figured it was a bad move because Qh8xRe8+.

It may buy some time and damage white's immediate attacking chances, but it's basically throwing the game.

Taking the pawn on a7 ensures white keeps the queen, but if black had followed up with Ng8-e7, white hasn't really accomplished much except taking a pawn and black develops his final knight and pairs up his rooks.

White picks up at least 2 pawns almost immediately at no cost (with more pieces to come), while keeping blacks king on the run, versus trading a queen for 2 rooks and turning the initiate over to black.