This is a (self-contained) followup question to https://math.stackexchange.com/questions/380672/analogue-of-the-schwartz-zippel-lemma-for-subspaces.
Let $f : \mathbb{R}^n \to \mathbb{R}$ be a nonzero multivariate polynomial of total degree $d$ over a field $\mathbb{R}$, and $S \subset \mathbb{R}$ be finite. Pick a positive integer $k$, choose $y_1, \ldots, y_k$ randomly and uniformly from $S^n$, and consider the $k$-variable polynomial
$$g(t_1,\ldots,t_k) = f(t_1 y_1 + \cdots + t_k y_k)$$
We seek an upper bound on the probability that $g(t)$ is the zero polynomial. Setting $t_i = \delta_{ij}$ for each $j$ and applying the Schwartz-Zippel lemma, we have
$$\Pr(g(t)=0) \le \frac{d^k}{|S|^k}$$
However, this bound does not use the fact that we have an entire subspace on which to be nonzero.
Question 1: Is there a slightly stronger unconditional bound when $k > 1$?
My reason for hope here is that the Schwartz-Zippel bound is achieved by a univariate polynomial with $d$ distinct roots, but in the subspace case $g(t) = 0$ only if we pick the same root for all $y_i$, giving $$\Pr(g(t)=0) \le \frac{d}{|S|^k}$$ if $f(x)$ depends on only one variable. In this case, we might save up to a factor of $d^{k-1}$.
Question 2: Say $f(x)$ is irreducible over $\mathbb{R}$, and we condition on $y_i$ being full rank. Is there an even stronger bound when $d > 1$, ideally growing in strength with $n$?
Edit: Question 2 is false as stated, because $y_i$ might be full rank but lower rank when projected onto the variables on which $f(x)$ depends. I am still hoping it holds if a suitable "depends on all variables" condition is added, but am not sure how to phrase this.