I know everything depends on a particular position, but in general would you prefer to:

a) Be one pawn down with 2 isolated king-side pawns

b) Exchange your Queen for a Rook and a Knight?

Try show it on board, so we have some better idea what you are talking about.

It's just a general question. With both options you effectively go 1 point down. But in a game of your own would you take the poor queen exchange to save a pawn + pawn structure?

Sometimes you see master games where this queen exchange happens. In my own games I would terrified of doing it (queen for rook + knight). But the question is should I be more afraid of that option than losing a pawn and messing up the structure?

Perhaps my fear of losing the queen is given too much weight? Perhaps I sometimes need to embrace this exchange to maintain chances? Or is it generally a crazy idea? I really don't know.

So, have you done it, and would you do it? Sorry if it's just a stupid question...

It's not stupid, it just depends on many other facts.

I can see, if I play vs someone more than few games, if he values his Queen to much. If he avoid Queen exchange more than once, I know he values his Q, and that he is somewhat lost when exchange happens. So I try to make it happen as soon as I can. Queen can be very dangerous alone, but If you play correct, and you are R+N vs Q, Queen cant exchange for either of that 2. It can only try to win one for free.

Well, depending on the position the pawn structure change could even be a good thing. If he can't attack the weakness it isn't a weakness, and it might give you some open line you can use.

Similarly it rather matters if the rook and knight I get I have while he has a queen have good prospects or not, same about his queen.

I'd say a queen down for a Rook and a Knight, as in some positions (for example pawns on one wing without blatant weaknesses), the side with the minors might even be on top, while in the first case, you're just statically worse without compensation.

But of course, it depends on the position.

This is quite an interesting question, even if it is dressed up a bit oddly by the OP.

Generally speaking, I do think it's preferable to be material down but with the material imbalanced than straight material down. Material imbalances give much better chances of confusing the game.

As long as i have both of my knights on the table I am confortable to give up my queen 

The question appears to be a purely theoretical one since in a game other factors will matter, and you will know other information. The closest to an answer I could give would be to see what the statistics are for games where one or other situation has occurred. I am too lazy to do the searches necessary to get the best answer (as my database does not allow me to specify that white has one pawn less than black, only the relative material and constraints on the numbers of each piece for each side, etc., so I would have to do 7 searches for 8 v 7 pawns, 7 v 6 pawns, etc, with material difference -1 in each case).

But a fudge is to search for games with white having h, f, but no g, e pawns and one pawn down in material. In most of these, the one pawn down will probably actually be a pawn rather than R v B+N for example.

In these (c. 200,00 games) white managed 40% (must have been a lot of good gambits, even those I have excluded king's gambit by saying white has h and f, rather than e and g (should also do these if I wasn't so lazy).

For the alternative, again a thorough job would require several searches because of the lack of an option like "white has one more rook than black", but to give an indication, I filtered all games where black has a queen, a rook and a knight and white has no queen, two rooks and two knights, with the material difference being -1 (the database uses the rough and ready 9/5/3/3/1 material count). In these 1700 games, white only manages 37%.

So I would say given no other better choice (such as chewing one's own foot off), the preferable option would be to be a pawn down. The difference was only 3.5%, but probably statistically significant with those numbers of games. I have given enough info for someone with half an hour spare to do this properly with between 10 and 20 searches, or using a more versatile database. The answer might even be a useful hint in real games, supporting my belief that being a queen down for a rook and knight requires quite a bit of compensation.

Perhaps a clearer indication of how big a disadvantage it is in the ending is that endings with Q v R+N with equal pawns  (no other pieces) scored over 75% for the queen.