If you want all your nodes to have the same configuration, you could define a polymorphic pivot on each node with the autoStart property set to false (please refer to http://support.quartetfs.com/confluence/display/AP4/Distribution%27s+properties#Distributionsproperties-autoStart). However, you'll then have to start one of them: this can be achieved by using the monitoring beans for instance.
Another possibility would be to deploy your horizontal nodes "as usual", then start an additional "proxy" node which will participate to both horizontal and polymorphic clusters. This extra node does not need any data: it's basically an empty horizontal pivot with a polymorphic pivot on top of it.