As M. Kanté pointed out, it's open whether or not graph isomorphism is FPT when parameterized by tree-width. Furthermore, I don't believe there is any complexity-theoretic barrier to creating an FPT algorithm in this case.
For a survey of what's known about the fixed-parameter tractability of graph isomorphism, see the introduction of my paper with Anuj Dawar and Eryk Kopczyński here. In the paper we show graph isomorphism is FPT in the tree-depth of a graph, which is a necessary (but not sufficient) condition for graph isomorphism to be FPT in tree-width.