That is not an answer. I want to give an example where the argument of Erdős does not work directly. Consider an action of group $\Gamma$ on $\mathbb R^3$ generated by the reflections in two lines $x=z=0$ and $x+1=z=0$ and a glide rotation $(x,y,z)\mapsto (-y,x,z+1)$. The reflections in $x=z=0$ and in $x+1=z=0$ generate two nonconjugate maximal $\mathbb Z_2$-subgroups. (There are more nonconjugate maximal subgroups, but the total number is still $<2^3$.) These two groups corespond to two singular circles, say $\Sigma_1$ and $\Sigma_2$ in the factor $X=\mathbb R^3/\Gamma$. Let us try to mimic argument of Erdős. Take subsets $X_i$ of $X$ of midpoints $m$ between $x\in X$ and a closest $x_0\in\Sigma_i$ to $x$. As in the argument of Erdős we have* $\mathrm{vol}\\, X_i>\tfrac{1}{2^3}\cdot\mathrm{vol}\\, X$. BUT $X_1\cap X_2$ has interior points and here argument brakes into parts. **Comment.** (*) Since fixed point sets are 1-dimensional, it would be enough to take $m\in [xx_0]$ such that $\tfrac{|mx_0|}{|xx_0|}=\tfrac1{2\sqrt[3]{2}}$. But even in this case one has interior points in $X_1\cap X_2$ (the borderline in this example seems to be $\tfrac13$).