2 Rooks vs Queen

The "idea itself" is fine as long as the simplification is winning for you... because otherwise, as you said yourself, it's a good knight for a not so good bishop.

I think the most important thing to note here is after 31...Qxb2 let's pretend black gets to move again, what does he want? Well white's pieces are loose, maybe he wants to attack one, and in general he wants to push his queenside. But look, the queen can't attack an undefended piece and the bishop is blocking the queenside pawn advance.

So would I contemplate NxB? Yes I would, but not on that particular move. I'd probably play e6 and make black burn a tempo on whatever he's going to do, then maybe I'll take it next move. Your d1 rook and knight look loose, but they're sort of defending each other pretty well. After e6 black has no time for Ba4 or Be2, and the queen can't pressure anything either.

The idea of "my move improved my position, can you make a move that improves yours?" can be applied to a lot of different situations. And when they can't improve their position, usually you don't want to make any changes (like opening lines or making trades or changing the pawn structure). You want to let them play as many not useful moves as possible. After the black queen recaptures on b5 black has useful moves. That's how I see it.

And yeah, time trouble is tough. At least for the sake of analysis it led to an interesting position

Just going on general considerations, keeping the queen and taking the g6 pawn looks about as good to me.  You need to dive in to specifics and think about how the trade impacts the d5 pawn and your own epawn and other issues to realize the two rooks are the much better choice.  Its a tactical issue as much as a positional one.  You made the right call, but I'm not sure I'd get there in a blitz game.  The two rooks are often the more complicated choice given that maintaining careful coordination is often critical.  

The two rooks are usually stronger than the queen, and this game drives home that point, though your opponent could have played better.  For example, 35. ..b6 was a critical waste of tempo, if he pushes the a-pawn instead, then that pawn gets to a2 by the time Rd8 is played, and so he can queen that pawn and effectively trade it for the e-pawn.

That said, once someone trades the queen for two rooks, the side with the queen should be looking for a perpetual check, and he'll need the bishop in order to sacrifice it for the pawns in front of your king.  So instead of 31. ..Qxb2? correct is 31. ..Bd7! with Qa4 to follow, then black looks for a chance to sacrifice the bishop with Bxh3 to open up the white king.

In general the important squares to control are the squares in front of and behind the passed pawns (both white's and black's). It looks like sometimes the engine is liking black because he's able to put the queen on e2 and it's hard for you to control the square in front of your passed pawn to escort it down the board.

As for why specific moves do and don't work it's hard to say. A lot of calculation is necessary. I'd definitely want tons of time on my clock for this endgame.

Well its a good thing that you traded your queen for two rooks because in this position it is the best practical choice. I dont care if the engine thinks it is a draw, for a human the advanced passed e pawn offers very good chances. GG.