I think that YangMills is probably right that Donaldson never wrote the conjecture down. But there are some interesting circles of ideas surrounding the conjecture which deserve mention which, again he probably never wrote down, but I think motivated some of his work on symplectic manifolds: namely, the idea that one could define invariants of symplectic manifolds inductively by dimension. For instance, take a 4-dimensional symplectic (Donaldson) hypersurface in a symplectic 6-manifold. There is a sense (only an asymptotic sense) in which you can do this uniquely. Is that enough to use smooth invariants of the 4-manifold to define symplectic invariants of the six-manifold? No-one has ever succeeded, due to the complicated nature of the asymptotic uniqueness.
The question about 4/6-manifolds which Chris Gerig is asking about is probably motivated by a more concrete phenomenon: smooth (i.e. Seiberg-Witten) invariants of symplectic 4-manifolds see the same information as symplectic (i.e. Gromov-Witten) invariants; after crossing with a sphere, homeomorphic but non-diffeomorphic symplectic 4-manifolds become diffeomorphic 6-manifolds, however symplectically you can still detect their Gromov-Witten invariants by counting curves in the 6-manifold (see the early papers of Ruan). The classic example is to compare the Barlow surface (a surface of general type) and a (homeomorphic) blow-up of the projective plane. One is minimal, the other has many -1-curves and you can still see these after crossing with a sphere.
This also explains why 4 and 6 are the relevant dimensions: smooth geometry in dimension 4 and symplectic geometry in dimension 6 are both "hard" in the Gromov sense. There are elliptic PDEs whose moduli spaces can be used to distinguish exotic pairs. By contrast there's no hard smooth invariants for 6-manifolds, so the question doesn't generalise.
I guess the conjecture Chris mentions is the most optimistic extrapolation of this observation, designed to encourage people to think about the circle of ideas.