Mrthdrb

Probably this question can be phrased in a much greater generality, but I will just state it in the generality I require. I work over $mathbbC$.

Let $C_1, C_2 subset mathbbP^1$ be non-empty open subsets and $f: X to C_1 times C_2$ a non-trivial finite etale cover. Does there exist $iin 1,2$ such that the composition $X to C_1 times C_2 to C_i$ has non-connected fibres?

The answer is no, at least in general, as shown by the following counterexample.

Take a double cover $barf colon barX to mathbbP^1 times mathbbP^1$, branched over a reducible
curve of the form $B=L_1 + L_2 + M_1 + M_2$ (here $|L|$ and $|M|$ are the two pencil of lines on the quadric).

Such a cover exists because $B$ is $2$-divisible in $mathrmPic(mathbbP^1 times mathbbP^1)$, and corresponds to an étale cover $f colon X to C_1 times C_2$, where each $C_i$ is $mathbbP^1$ - two points.

If these points are (say) $0$ and $1$ in both factors, then the equation for $X subset mathbbC times (mathbbC-0, , 1)^2$ is
$$z^2 = xy(x-1)(y-1), quad f(z, (x, ,y)) = (x,, y).$$

It is clear that the general line in $|L|$ and $|M|$ intersects the branch locus $B$ transversally at two points, hence both compositions $$barX to mathbbP^1 times mathbbP^1 to mathbbP^1$$ have connected fibres, the general one being isomorphic to $mathbbP^1$ (double cover of $mathbbP^1$ branched at two points).

Then both compositions $$X to C_1 times C_2 to C_i$$ have connected fibres, the general one being isomorphic to the $mathbbP^1$ above minus the ramification, i.e. $mathbbP^1$ minus two points, that is clearly connected.

In the same vein, choosing as $B subset mathbbP^1 times mathbbP^1$ a divisor of type $$B = sum_i=1^2g+2 L_i + sum_i=1^2g+2 M_i,$$
both compositions $$barX to mathbbP^1 times mathbbP^1 to mathbbP^1$$ have connected fibres, the general one being isomorphic to a hyperelliptic curve $Sigma_g$ of genus $g$, and so both compositions $$X to C_1 times C_2 to C_i$$ (here each $C_i$ is $mathbbP^1$ minus $2g+2$ points) have connected fibres, the general one being isomorphic to $Sigma_g$ minus $2g+2$ distinct points.

The question already has a beautiful answer, but here's a different point of view which you may find helpful.

Let $F_i = pi_1(C_i, x_i)$, which is a free group on $#(mathbfP^1setminus C_i) - 1$ generators (the etale fundamental group will be the profinite completion of this). Then $F = pi_1(C_1times C_2, x_1times x_2) = F_1times F_2$.

A finite etale cover of $C_i$ or $C_1times C_2$ corresponds to a finite set (its fiber at the basepoint $x_i$ or $x_1times x_2$) with an action of $F_i$ or $F$. The cover is connected if and only if the action on that set is transitive.

Let $sigma_i colon C_ito C_1times C_2$ be the section $sigma_1(x) = (x, x_2)$, $sigma_2(x) = (x_1, x)$. Then for a finite etale cover $Xto C_1times C_2$, the composition $Xto C_1times C_2to C_i$ has connected fibres if and only if the pull-back of $X$ along $sigma_2-i$ is connected.

So now the question is equivalent to: suppose that $S$ is a finite set with more than one element with an action of $F=F_1times F_2$. Is it possible that $F_1$ and $F_2$ both act transitively on $S$? It is very easy to construct such examples.

The easiest one could be $S$ with two elements, with every generator of each $F_i$ acting by a nontrivial involution. If there are only two punctures on each curve, this coincides with Francesco Polizzi's construction, and we see that his example is in some sense minimal.