Edit: My previous alleged counter-example was flawed as pointed out in the comments.
I'll leave the first part as a partial answer.
If one can show that the entropy is finite, then the entropy must, in fact, be zero as shown below.
Suppose $T$ has finite entropy. We will use the Variational Principle as you intended and a property of measure-theoretic entropy.
Let $\mathcal{M}$ be the set of all probability measures on $X$ preserved by $T$. Then $h(T) = \sup_{\mu \in \mathcal{M}} h_{\mu}(T)$, where $h_{\mu}(T)$ is the measure-theoretic entropy w.r.t. $\mu$.
Let $A_N = \bigcup_{n=0}^N P_n$, where $P_n = \{x \colon T^nx = x\}$ as you defined.
Since $X = \bigcup_{N \geq 0} A_N$ and since $A_N$ is a nested increasing sequence, we know for any $\mu \in \mathcal{M}$ that $\lim_{N \rightarrow \infty} \mu(A_N) = 1$.
In particular, given any $\mu \in \mathcal{M}$, the set $A_N$ has non-zero measure for all sufficiently large $N$.
Also, each $A_N$ is invariant.
Here is a property of measure-theoretic entropies: denote by $\mu_{A_N}$ the conditional measure on $A_N$ and likewise $\mu_{X\backslash A_N}$.
Then
$$ h_{\mu}(T) = \mu(A_N) h_{\mu_{A_N}}(T|_{A_N}) + (1-\mu(A_N)) h_{\mu_{X\backslash A_N}}(T|_{X\backslash A_N}). $$
You know that $h_{\mu_{A_N}}(T|_{A_N}) = 0$ for all $N$ because $T|_{A_N}$ is periodic and that
$$ h_{\mu_{X\backslash A_N}}(T|_{X\backslash A_N}) \leq h(T|_{X\backslash A_N}) \leq h(T). $$
Thus,
$$ h_{\mu}(T) \leq (1-\mu(A_N)) h(T). $$
Taking the limit $N \rightarrow \infty$, we conclude $h_{\mu}(T) = 0$ for any $\mu \in \mathcal{M}$ and, hence, $h(T) = 0$ as desired.