I agree with everyone above, but I would add that the true answer to your question is never. Even Feliks messes up on his lookahead and pauses at times. These are normally due to lockups, but still you honestly can never be too good at look ahead. I still work on it every day and I am working my way into the 10's now. IMO, its not something you ever master, but rather something that you get to a point where its no longer your focus or even really a thought. Eventually you will start to lookahead more than one pair at a time as well so you truly never end or master it. But as people mentioned above, the key is to start looking for your next pair the moment you find your current one. Also, I tend to keep a mental map of where any of my cross colors (the f2l corners) that I see and their edges if possible and that way finding the next pair is easy because you either know exactly where it is or you only need to look for the edge. My goal right now is to know where all of the corners are by the end of second pair that way I can do the last two with little thought and start working on edge control and OLL case reduction.
If your curious how I do it you can check my vids in my sig. I have a multi-part tutorial on looking ahead.
Edit: Oh and you mentioned you feel the urge to go faster or even im guessing the fastest you can during f2l and there are three main problems I see with this at the stage your at: first of all your lookahead likely isnt fast enough to track the pieces when you are moving at full speed which means you will have to a lot pause more and even if you only do it lets say twice during cross and f2l, thats at least two seconds in which you were literally achieving nothing PHYSICALLY on the cube. So in terms of solving those two seconds are a complete waste of time (not literally because you are still thinking, but you actually achieve nothing on the cube itself).
Second, in my experience it promotes bad habits, laziness and contentment when solving. By this I mean a lot, not all, but a large amount of the people who i see trying to go full speed tend to focus on trying to move faster to get better as opposed to learning better algs, working on pair choice, doing better and more x-crosses or any of the other far more important things to cubing than just tps. Also, because you get so used to going at full speed most people who do this don't like learning new algs because it slows them down in the short run even though it will be better in the long run. I know people who are practically sub 10 and they still have cruddy to average f2l algs, but they don't/wont/are too lazy to change them because it will make them slower now even though if they did they would probably start dropping time like crazy in the long run. Unless we are talking about sub 9 times you really dont need to have that fast of tps; I can get sub 10's going at maybe half speed (3-4 tps during f2l), and I can also get them going at full speed (6-8 tps during f2l); I go full speed during OLL/PLL btw. If you go slower you will be more tempted to optimize your solving method and focus on far more important things like lookahead.
Finally, you should never move faster than your mind can comprehend, especially during F2l and if you are going full speed then you pushing or surpassing your mental limits which means you are missing out on so many opportunities on the cube including pair preservation, multislotting, next pair influencing and soo much more. But since you are going so fast your mind doesn't have enough time to track everything and so it just focuses on the easier and most trained solutions and doesn't have the time for any real thinking, adaption or innovation. Again this is just based on what I have seen in teaching my class so take from it what you want, but imo full speed tps is really not that important until you reach sub 10.