Let us take $G=Gl(n,\mathbb{C})$(considered as a linear algebraic group). Let us take $x \in G$: we know that its orbit $\mathcal{O}_x$ under the coniugation action is isomorphic (as algebraic varieties) to $$ G/C_G(x) \cong \mathcal{O}_x $$ via the action map where $C_G(x)$ is the centralizer group.In particular the action map realizes $\mathcal{O}_x$ as the base space of a principal $C_G(x)$ bundle.

I was trying to determine what happens if we take the following more general situation: we take $$n_1+ \cdots n_k=n $$ and we fix partitions $\lambda_1,\dots \lambda_k$ of $n_1,\dots n_k$ respectively. To each such a partition we can associate in a unique way an (uppertriangular) unipotent Jordan matrix , in the canonical way, which we denote $J_{\lambda_i}$.

We then call $U_{\lambda_1,\dots \lambda_k}$ the set of elements of $Gl(n,\mathbb{C})$ with $k$ different eigenvalues such that the Jordan form of the $i$th eigenvalue is given by $J_{\lambda_i}$.

If we denote $$W=\{ (x_i) \in \mathbb{G}^k_m \ | \ x_i \neq x_j \ ,\ i \neq j\} $$ we have a map $$W \times Gl(n,\mathbb{C}) \to U_{\lambda_1,\dots \lambda_k} $$ which sends the element $((x_i),g) $ to $g $ applied to the block upper triangular matrix with blocks given by $x_iJ_{\lambda_i}$.

The morphism is surjective and so $ U_{\lambda_1,\dots \lambda_k}$ is constructible: I've tried to prove some of its geometrical properties but I didn't get much. This morphism in general does not seem a $C_G(x) $ principal bundle to me:as the fiber can happen to be bigger than elements of the form $$((x_1, \dots x_k),h) $$ with $h$ in the associated centralizer. Is there a way to change the approach and make this variety a base of a principal $C_G(x)$ bundle?

I was not even able to prove whether this variety is smooth or not ( apart from the regular semisimple case $\lambda_i=1$ for example or other degenerate ones). Is there a reference in the literature for this?