Bell's theorem places limits on the correlations that a theory like this can explain. Here's a simple example of a Bell violation:
1) Alice and Bob get together and conspire. They can do anything they want here.
2) Alice and Bob are then taken to separate interrogation chambers, to be quizzed by separate interrogators.
3) Each interrogator randomly (and independently) chooses one of three yes-or-no questions to ask their subject. Both interrogators are working from the same list of 3 questions. Alice and Bob are asked the questions and the questions and their answers are recorded.
4) This whole process is repeated many times. When the results are tabulated, it's found that every time Alice and Bob were asked the same question, they gave the same answer. But when they were asked different questions, 75% of the time they gave different answers.
How could Alice and Bob have achieved this? Classically, they couldn't. In order to always give the same answer to the same question, they'd have to (either directly or indirectly) agree beforehand on which answer to give to which question.
Since there are three possible questions but only two possible answers to them, any strategy they pick must give the same answer to (at least) two of the three questions. Therefore, if their interrogators pick different questions there's at least a 1/3 chance they'll pick the questions with the same answer.
But with quantum entanglement, the chance of getting the same answer from both is can be only 1/4. Something weirder is going on.
(BTW, for completeness, here's how the quantum results I described can be achieved: Alice and Bob are replaced by electrons, prepared in the (entangled) singlet state. The "interrogations" are measurements of their spin (up or down) on the 0°, 120°, or 240° axes, with one of the detectors set up upside-down. This is not the way it's actually done, but it's easier to explain & see the paradox this way.)
Give one to a friend, walk to the ends of the earth, toss the coin, then call your friend: comparing notes, they've come up different again... spooky!
Maybe the coins colluded when they were made, and there's no spooky action at a distance, just some hidden information inside the coins. This is what the Bell tests disprove. You can make a series of measurements that result in data that could not have been agreed upon by the coins ahead of time.
(In quantum mechanics, the entangled particles are never perfectly correlated or anticorrelated like this, but they coincide or not at a rate much different than ordinary particles)
Very roughly speaking, this picture is actually flawed. Charlie can not choose the fruit to eat in quantum mechanics. In this picture you would also transmit Charlie's choice to Alice faster than the speed of light which is not possible in quantum mechanics. The important part of this picture is that Bob does not leave the room with a predetermined fruit, the choice happens later outside of the room but Alice still always ends up with the other fruit, Alice and Bob never end up with the same fruit.
> "Charlie would ask Bob to eat the pear"
This seems to be the key part. So when they run experiments on this, a certain result can be forced?
> "The important part is that Bob does not leave the room with predetermined fruit."
Right, is this aspect measured somehow or only inferred from the theory?
Have I look at my other comment with coins in boxes [1], that picture captures the situation better and in a way that is not easily translated to fruits.
The problem is that allowing Charlie to make the choice allows him to use entanglement to transmit information faster than the speed of light.
No, a result can't be forced. The analogy being used isn't very close. I'm going to summarize an explanation used in (IIRC) The Elegant Universe that may be helpful.
Mulder and Scully each get shipped a box that's entangled with the other's box. These boxes have three windows. You're allowed to open only one window, at which point you'll see a red or green light. These boxes further have the property that they're entangled: if Mulder and Scully open the same window, they'll see the same color.
Mulder and Scully open the same window on each box, and sure enough, they see the same color. Mulder claims that it's spooky action at a distance, Scully claims that the boxes are simply preprogrammed with a preset color pattern: one of "red, red, green", "green, red, green", "red, red, red", and so on.
Mulder thinks for a bit and devises an experiment. They now acquire a large number of these boxes, each paired with a box in the other's possession (and in such a way they can identify which of their two boxes correspond to one-another). They proceed to open doors randomly, recording the results. Now they compare notes.
In the case where they both opened the same door on an entangled pair of boxes, they always get the same result. So far so good. But what happens when they look at the results as a whole? Well, Scully's theory was that these boxes are preprogrammed and the data we've collected falsifies this theory.
The only possible patterns, ignoring specific color and order, are ones where two windows reveal the same color and one window reveals another color (any ordering of GGR or RRG) or where all three windows open to the same color (GGG or RRR). In the latter case, their results will obviously match 100% of the time. In the former case, 2/3 of the time their results will match 2/3 of the time (for 4/9ths) and 1/3 of the time their results will match 1/3 of the time (for an additional 1/9th), adding up to 5/9ths.
So if Scully's theory is true, when opening doors at random, they should have recorded the same result at least 5/9ths of the time (more if some boxes are programmed with all windows revealing the same color).
Unfortunately for Scully, their records show that their results matched exactly 50% of the time. No matter how complicated your "programmed" hidden variable theory is (pseudorandom generators with the same seed, etc.), it is fundamentally disproven if the observed match rate is less than 5/9ths. The simplest explanation that fits this experimental result is that both boxes choose to display a color truly randomly when a given window is opened, but they both arrive at the same random color when the same window is opened!
Again, this statistical argument works against all hidden variable theories.
Edit: I like danbruc's answer better.
1) Alice, Bob, and Charlie are in a room.
2) Alice has only an apple and a pear, seen by Charlie.
3) Charlie leaves the room.
4) Alice gives one of the two fruits to Bob.
5) Bob leaves the room to meet up with Charlie and starts eating the pear.
6) Charlie instantly knows that Alice has the apple, no matter where she has ended up in the universe (which is not spooky at all).