This is probably not yet a final answer but may shine some additional light on the problem:
For simplicity, I assume that $L$ is finite-dimensional and defined over the reals (for some other field of char $0$, the following should still work).

The symmetrization map can be viewed as a linear map
\begin{equation}
\sigma\colon \mathrm{S}(L) \longrightarrow \mathrm{U}(L)
\end{equation}
from the symmetric algebra over $L$ into the universal envelopping algebra. It is now possible (essentially via PBW) to show that this is a fitration compatible linear bijection. Thus it allows to pull-back the product of $\mathrm{U}(L)$ to $\mathrm{S}(L)$. The result is the star product of Gutt / Drinfel'd (both in 1983, I guess). A further canonical isomorphism yields that the symmetric algebra is nothing else than the poylnomials on the dual $L^*$ (suppose $L$ is finite-dimensional for convenience) Thus your question is equivalent to the following task:

What is the Gutt star product commutator of two (homogeneous) polynomials on $L^*$?

Gutt has computed many properties of this star product and fan almost explicit formula. However, it essentially involved the full BCH series of $L$, so my guess is that a complete answer might be as complicated as computing BCH.

The Gutt star product can be characterized nicely as follows: take $x, y \in L$ and view them as linear polynomials on $L^*$ as usual. Then form the *formal* exponential functions $e_{\hbar x} (\alpha) = \exp(\hbar \alpha(x))$ and similarly for $e_{\hbar y}$ where $\hbar$ is your formal parameter. Then $\star_{\mathrm{Gutt}}$ is uniquely determined by
\begin{equation}
e_{\hbar x} \star_{\mathrm{Gutt}} e_{\hbar y}
=
e_{\mathrm{BHC}(\hbar x, \hbar y)}
\end{equation}
I hope I got the signs right :) You can find this formulas also in Section 8 of q-alg/9707030 (published in Commun. Math. Phys.). There are many more papers on the Gutt star product, so a little MRsearch will probably give some addition info.

The solution of your problem is now obtained by differentiating the above equation with respect to $\hbar$ sufficiently often and use polarization afterwarts. But as I sad, you need to know BCH quite well to efficiently do that. In the end you take commutators\ldots