Suppose $f(x) \mid f(x^2 - 2)$. Ghartal's argument has already shown that if $x$ is a root of $f$, then $x^2 - 2$ is also a root of $f$. In fact, by the same argument, we have that for all $x \in \mathbb{C}$, the multiplicity of $x^2 - 2$ as a root of $f$ is at least the multiplicity of $x$ as a root of $f$.
I claim that the above condition is also sufficient to have $f(x)\mid f(x^2 - 2)$. Indeed, we may write $f$ as
$$
f(x) = k\prod_{c \in \mathbb{C}} (x - c)^{m_c}
$$
for some $k \in \mathbb{C}$ (w.l.o.g. $k \neq 0$) and where $m_c \in \mathbb{N}$ is non-zero for finitely many $c \in \mathbb{C}$. Then
$$
f(x^2 - 2) = k\prod_{c \in \mathbb{C}} (x^2 - 2 - c)^{m_c} .
$$
Now by hypothesis we have $m_{c^2 - 2} \geq m_{c}$ for all $c \in \mathbb{C}$. Hence $(x -c)^{m_c} \mid (x^2 - 2 - (c^2 - 2))^{m_{c^2 - 2}}$ for all $c \in \mathbb{C}$. In particular, $f(x) \mid f(x^2 - 2)$. This proves the claim.
To find all degree three polynomials (in $\mathbb{R}[X]$, or $\mathbb{C}[X]$ if you wish) with the given property, it suffices to "trace back" the function $x \mapsto x^2 - 2$ three steps and then take products of the induced linear polynomials; respecting the claim above. The function $x \mapsto x^2 - 2$ does:
$$
+/- \sqrt{3} \mapsto 1 \mapsto - 1 \qquad \text{and} \qquad 0 \mapsto -2 \mapsto 2.
$$
so the possible polynomials of degree at most three are (up to multiplying by a constant):
- $(x + 1)$
- $(x - 2)$
- $(x + 1)^2$
- $(x - 2)^2$
- $(x+1)(x-2)$
- $(x-1)(x+1)$
- $(x+2)(x-2)$
- $(x+1)^3$
- $(x - 2)^3$
- $(x - 2)(x+1)^2$
- $(x - 2)^2 (x + 1)$
- $(x-1)(x+1)^2$
- $(x + 2)(x - 2)^2$
- $(x - \sqrt{3})(x-1)(x+1)$
- $(x + \sqrt{3})(x-1)(x+1)$
- $x(x + 2)(x - 2)$
- $(x + 2)(x - 2)(x + 1)$
- $(x - 1)(x + 1)(x - 2)$
The polynomials of degree exact three are the last 11.