This is not true in general. Indeed, let $X$ be a zero-mean random variable (r.v.) such that $Ee^{tX}<\infty$ for $t\in[0,\tau)$ but $Ee^{\tau X}=\infty$. Then for all $t\in(0,\tau)$ the left-hand side of the conjectured inequality will be finite but the right-hand side of the inequality will be $\infty$. 

---

It is quite easy to modify the above example by approximation so as to make the r.v. $X$ bounded. E.g., let $X:=X_n:=\min(Y,n)-\mu_n$, where $Y$ is an exponential r.v. with mean $1$ and $\mu_n:=E\min(Y,n)=1-2 e^{-n} n-e^{-2 n}\to1$ (as $n\to\infty$). Then the r.v. $X$ is bounded and 
$$K(t)=K_n(t)=\ln\frac{t e^{(t-1)n}-1}{t-1}-\mu_n t$$
for real $t\ne1$, with $K(1)=K_n(1)=\ln(n+1)-\mu_n$. So, letting $n\to\infty$, for real $t\in(0,1)$ we get $K(t)=K_n(t)\to\ln\frac1{1-t}-t$ and $K'(t)=K'_n(t)\to\frac t{1-t}$, so that the left-hand side of the conjectured inequality goes to a finite limit, where as the limit of the right-hand side of the conjectured inequality for $\tau=1$ is $\lim_{n\to\infty}tK_n(1)=\lim_{n\to\infty}t(\ln(n+1)-\mu_n)=\infty$. So, for each $t\in(0,1)$ and all large enough $n$ the conjectured inequality will fail to hold.