If $f: X\to Y$ is a separable map between varieties that is a bijection on closed points, is it true that $f$ remains separable when restricted to an integral subscheme $Z\subset X$?
If we drop the condition that $f$ is a bijection, the only relevant example I know is the quasielliptic fibration $\{y^2=x^3-t\}\to \mathbb{A}^1$ that maps $(x,y,t)$ to $t$, where we are working over a field of characteristic 3. If we restrict to the locus $y=0$, then we get the Frobenius map $\mathbb{A}^1\to \mathbb{A}^1$.