I don't know of a cross-platform SQL-oriented toolkit for this. You could consider having each node write changes to a uniquely-named file, and then using existing file-replication techniques to make sure they end up where they're going...