Chaudhary and Gordon ("Tutte polynomials for trees," J. Graph Theory 15, no. 3 (1991), 317-331) construct a couple of invariants that look very similar to yours. They prove that these invariants do in fact determine a rooted tree up to isomorphism.
Update: I think the answer to your original question is no.
The relevant invariant from the Chaudhary-Gordon paper is what they call $f_p(T;t,z)$. This is a polynomial in two variables $t,z$ that
satisfies the recurrence
$$ f_p(L(T);t,z) = t(z+1)f(T) + 1 - tz,$$
$$ f_p(T_1*\cdots*T_r;t,z) = f(T_1)\cdots f(T_r)$$
where $L$ means leafing and $*$ means grafting. (These are Prop
4(b) and and Prop 5 in Chaudhary-Gordon.) If I'm doing the algebra
right, your invariant is $P_T(z) = f_p(T;z+1,0).$
Chaudhary and Gordon give an example of two rooted trees on 8 vertices with the same values of $f_p(T;t,z)$. The edge sets could be labeled as
01,12,24,13,35,56,57 and 01,12,13,34,35,56,67, with 0 the root vertex in
both cases. (Probably a good idea to confirm this if you have code to
compute your invariant quickly.)
We can see any tree $T$ as the Hasse diagram of a poset whose smallest element is the root. I will freely identify the tree with the corresponding poset.
If we label by $i$ the $i$'th vertex added in the process, we get a rooted tree equipped with a linear extension, and each such equipped rooted tree occur with probability $\frac{1}{(n-1)!}$ (where $n$ is the number of vertices).
The number of ways to get a given rooted tree in this way is the number of linear extensions of this tree up to automorphisms of the rooted tree.
Hence the probability you're looking for is given by
$$\frac{L(T)}{|\text{Aut}(T)| (n-1)!}$$
where $L(T)$ is the number of linear extensions of $T$ and $\text{Aut}(T)$ is the automorphism group of $T$.
There are algorithms for computing the former in $O(n^2)$ (https://www.researchgate.net/publication/226291352_On_computing_the_number_of_linear_extensions_of_a_tree). I am not sure for the latter, but it can probably be computed efficiently using a variant of the AHU algorithm.
Best Answer
Late edit: having now read through the OP comments, I can see that my proof is essentially a carbon copy of @darij.grinberg's approach (although my derivation was independent). I'm okay to delete this answer once/if darij chooses to post theirs.
Pick a canonical ordering of unlabelled rooted trees, say, with lexicographical comparison of tuples $(n, T_1, \ldots, T_k)$, where $n$ is the number of vertices, $T_1, \ldots, T_k$ is the non-descending sequence of children subtrees. Throughout $T, T_1, T_2$ are unlabelled rooted trees in canonical form (that is, subtrees of any vertex are ordered as above).
Let $A_n$ be the set of pairs $(T, v)$ with $|T| = n$, $v$ is a non-root vertex of $T$. Also, let $B_n$ be the set of tuples $(T_1, T_2, k, u)$ such that $|T_1| + k|T_2| = n$, $u$ is a vertex of $T_2$. Observe that $|A_n|$ and $|B_n|$ are LHS and RHS of the recurrence in OP, more readily seen by rewriting $(n - 1)t_n = \sum_{m = 1}^{n - 1}mt_m \sum_{0 < km < n} t_{n - km}$.
The bijection between $A_n$ and $B_n$ is as follows: