There are two approaches to this - both giving the same result.
The first approach is what explains the term "derivative" (or "instantaneous rate of change") of a function "with respect to another function". Let's fix $x\in\mathbb R$. In some neighbourhood of $x$, the function $g(x)$ will be invertible, i.e. you will be able to express $x$ in terms of $t=g(x)=x^3-1$. (In this particular case, this is true on the whole $x$ axis, because $g(x)$ is $1-1$, but in other cases you may need to restrict the domain to possibly a smaller neighbourhood of $x$.) What you are asked to do is express $x=g^{-1}(t)$, then express $f(x)$ as a function of $t=g(x)$ and then calculate the derivative of that function $f(g^{-1}(t))$ (still as a function of $t$), and finally represent this derivative back as a function of $x=g^{-1}(t)$.
You could directly do this manipulation on given functions $f$ and $g$. If $t=g(x)=x^3-1$, then $x=\sqrt[3]{t+1}$ and $f(x)=x^2+16=(\sqrt[3]{t+1})^2+16$. Thus, the required rate of change is $\frac{df}{dt}=2(\sqrt[3]{t+1})\frac{1}{3}(t+1)^{-2/3}=\frac{2}{3}\sqrt[3]{t+1}\frac{1}{(\sqrt[3]{t+1})^2}=\frac{2}{3}x\frac{1}{x^2}=\frac{2}{3x}$.
However, let's apply the rules of calculus without rushing to substitute $f(x)$ and $g(x)$:
$$\frac{d[f(g^{-1})(t)]}{dt}=f'(g^{-1}(t))\frac{d[g^{-1}(t)]}{dt}=f'(g^{-1}(t))\frac{1}{g'(g^{-1}(t))}=\frac{f'(g^{-1}(t))}{g'(g^{-1}(t))}=\frac{f'(x)}{g'(x)}$$
The second approach is unjustified (or, rather, is justified by the first approach!) but is easy to memorize. You may have learned it in this form. The idea is: what you want to find "is" $\frac{df}{dg}$, so let's "extend" this fraction by "dividing" both the numerator and denominator by $dx$, which gives:
$$\frac{df}{dg}=\frac{df/dx}{dg/dx}=\frac{f'(x)}{g'(x)}$$
Now, with whichever approach, you can substitute the actual derivatives of $f(x)$ and $g(x)$, and the result is just $\frac{2x}{3x^2}=\frac{2}{3x}$.