I also did not get any 1/8, or used Taylor expansion, but...
Assume there is only 1 maximum. since that's the most extreme case for the bound, it should cover all other cases. Define $c$ the value in $[a,b]$ that achieves this. since $f$ is smooth, then if $c\neq a$ and $c\neq b$ then $f'(c)=0$.
Now we use MVT to find a bound on the second derivative.
$\exists z_1\in [a,c]$ where $f'(z_1)=f(c)/(c-a)$
$\exists z_2\in [c,b]$ where $f'(z_2)=-f(c)/(b-c)$
Therefore $\exists z_3\in [z_1,z_2]$ where
$$f''(z_3)=\frac{f'(z_2)-f'(z_1)}{z_2-z_2}\\
=\frac{f(c)/(c-a)+f(c)/(b-c)}{z_2-z_1}\\
=f(c)\frac{b-a}{(b-c)(c-a)(z_1-z_2)}\\
$$
To get a bound, we want to pick $c,z_1,z_2$ to minimize the right hand side, so we pick $c=(b+a)/2$, $z_2=b$, $z_1=a$. Simplifying gives
$f''(z_3)\geq 4f(c)/(b-a)^2\iff f(c) \leq \frac{(b-a)^2}{4}f''(z_3)$
I'm not quite sure where that extra factor of 2 is coming in since I think without more assumptions MVT is tight...