Dear Ines,
Let me supplement my first response with a related (possible) answer to your question which makes use of the expansion of power-symmetric functions in terms of Schur functions (an identity which you and I have discussed in other threads of this forum).
Given a partition $\mu= \big(\mu_1 \geq \dots \geq \mu_k > 0 \big)$ of $n$ with $k$ parts, you will recall that the corresponding power symmetric function $p_\mu$ is defined as the product $p_{\mu_1} \cdots \, p_{\mu_k}$ of elementary power-symmetric functions $p_{\mu_1}, \dots, p_{\mu_k}$
where
\begin{equation} p_r \ = \ \sum_{i \geq 1} \, x_i^r \end{equation}
for each positive integer $r$. The expansion of a power-symmetric function in terms of Schur functions is
\begin{equation} p_\mu \ = \ \sum_{|\lambda| \, = \, n} \, \chi^\lambda_\mu \, s_\lambda \qquad \qquad (\dagger)
\end{equation}
where the sum is taken over all partitions $\lambda$ of $n$ and $\chi^\lambda_\mu$ is the value of the irreducible character $\chi^\lambda$ of the symmetric group $S_n$ evaluated on the conjugacy class $C_\mu$. We'll begin by specialising the left and right hand sides of equation $(\dagger)$ using the values $x_i = q^{\, i}$. The power-symmetric case is a simple calculation involving geometric series and one checks that
\begin{equation} p_\mu \big(1, q, q^2, \dots \big) \ = \ \big(1-q \big)^{-k} \cdot \big[ \, \mu_1 \big]^{-1} \cdots \big[ \, \mu_k \big]^{-1}
\qquad \qquad (*)
\end{equation}
where $\displaystyle [r] = {\, \, \, 1 - q^{\, r} \over {1 -q}} $ is the combinatorist's $q$-number
of a positive integer $r$. The specialisation of the Schur function is the celebrated gem (of Stanley ?) with which you began the discussion, namely
\begin{equation}
\begin{array}{ll}
\displaystyle s_\lambda \big(1, q, q^2, \dots \big)
&\displaystyle = \ q^{\eta(\lambda)} \,
\prod_{\stackrel{\scriptstyle \text{boxes}}{b \, \in \, \lambda}}
\, \Big(1 \, - \, q^{\, \text{hook}\, (b)} \Big)^{-1} \\ \\
&\displaystyle = \ q^{\, \eta(\lambda) \, + \, H(\lambda)} \, \big(1 -q \big)^{-n} \,
\prod_{\stackrel{\scriptstyle \text{boxes}}{b \, \in \, \lambda}}
\, \big[\!\big[ \, \text{hook}(b) \, \big]\!\big]^{-1} \\ \\
&\displaystyle = \ q^{ {1 \over 2} \big( \eta(\lambda) \, - \,\eta(\lambda^{\text{op}}) \big)} \, \big(1 -q \big)^{-n} \,
\prod_{\stackrel{\scriptstyle \text{boxes}}{b \, \in \, \lambda}}
\, \big[\!\big[ \, \text{hook}(b) \, \big]\!\big]^{-1} \qquad \qquad (**)
\end{array}
\end{equation}
where $\displaystyle [\![ \, r \, ]\!] = {q^{\scriptstyle {1 \over 2}r} -
q^{ \scriptstyle -{1 \over 2}r} \over {q^{ \scriptstyle {1 \over 2}}} -
q^{\scriptstyle -{1 \over 2}}}$ is the physicist's $q$-number of a positive
integer $r$, where $\text{hook}(b)$ is the hook-length of a box in the young diagram of $\lambda$, and where
\begin{equation}
\begin{array}{ll}
\displaystyle
H(\lambda) \
&\displaystyle = \ {n \over 2} \, - \, {1 \over 2} \, \sum_{\stackrel{\scriptstyle \text{boxes}}{b \, \in \, \lambda}}
\text{hook}(b) \\ \\
&\displaystyle = \ - {1 \over 2} \Big( \eta \big(\lambda \big) \, + \, \eta \big( \lambda^{\text{op}} \big) \Big)
\end{array}
\end{equation}
The physicist's $q$ numbers are more convenient in this present context because $[\![ \, r \, ]\!] = r$ and $\displaystyle {d \over {dq}} \, [\![ \, r \, ]\!] = 0$ at $q=1$. Inserting formulae $(*)$ and $(**)$ into
equation $(\dagger)$ we obtain
\begin{equation}
\big(1-q \big)^{n-k} \cdot \big[ \, \mu_1 \big]^{-1} \cdots \big[ \, \mu_k \big]^{-1}
\ = \ \sum_{|\lambda| \, = \, n} \, \chi^\lambda_\mu \
q^{ {1 \over 2} \big( \eta(\lambda) \, - \,\eta(\lambda^{\text{op}}) \big)} \,
\prod_{\stackrel{\scriptstyle \text{boxes}}{b \, \in \, \lambda}}
\, \big[\!\big[ \, \text{hook}(b) \, \big]\!\big]^{-1}
\end{equation}
Upon differentiating by $q$, specialising $q=1$, and invoking the
(classical) hook-length formula we obtain
\begin{equation}
\begin{array}{ll}
\displaystyle - \delta_{k, n-1} \cdot \mu_1^{-1} \cdots \mu_k^{-1}
&\displaystyle = \ \sum_{|\lambda| \, = \, n} \, \chi^\lambda_\mu \ {1 \over 2} \Big(
\eta(\lambda) \, - \,\eta(\lambda^{\text{op}}) \Big) \,
\prod_{\stackrel{\scriptstyle \text{boxes}}{b \, \in \, \lambda}}
\, \text{hook}^{-1}(b) \\ \\
&\displaystyle = \ \sum_{|\lambda| \, = \, n} \, \chi^\lambda_\mu \,
\ {\chi^\lambda_{1^n} \over {2n!}} \,\Big(\eta(\lambda) \, - \,\eta(\lambda^{\text{op}}) \Big)
\end{array}
\end{equation}
where $1^n$ is the partition corresponding to the conjugacy class of the
identity element in $S_n$. Define
\begin{equation}
F(\lambda) \ := \ {\chi^\lambda_{1^n} \over {2n!}} \,\Big(\eta(\lambda) \, - \,\eta(\lambda^{\text{op}}) \Big)
\end{equation}
In view of Schur's second character orthogonality relations it follows that
$F(\lambda)$ is proportional to $\chi^\lambda_\varrho$ where $\varrho = 21^{n-1}$ is the partition corresponding to the conjugacy class $T$ consisting of all transpositions in $S_n$. Check that in fact
\begin{equation} F(\lambda) \ = \ -{1 \over 2} \,{|T| \over {n!}} \, \chi^\lambda_\varrho
\end{equation}
and so arrive at the following character-theoretic account of the $\eta$-value of a partition, namely:
\begin{equation} \eta\big( \lambda \big) \, - \, \eta\big( \lambda^{\text{op}} \big) \ = \ -|T| \, {\chi^\lambda_\varrho \over {\chi^\lambda_{1^n} }}
\end{equation}
yours, a. leverkühn