The recurse and recurseArchives notions in the VisitorAttributes have been replaced with a more general VirtualFileFilter recurseFilter. The filter controls whether recursion of non-leaf files is done:

public class VisitorAttributes
{
...
 /**
 * Whether to recurse into the non-leaf file<p>. If there is a recurse
 * filter then the result will be its accepts(file) value.
 *
 * Default: false
 *
 * @return the recurse flag.
 */
 public boolean isRecurse(VirtualFile file)
 {
 boolean recurse = false;
 if( recurseFilter != null )
 recurse = recurseFilter.accepts(file);
 return recurse;
 }

 /**
 * Get the recurse filter.
 * @return the current recurse filter.
 */
 public VirtualFileFilter getRecurseFilter()
 {
 return recurseFilter;
 }

 /**
 * Set the recurse filter.
 *
 * @param filter the recurse filter.
 * @throws IllegalStateException if you attempt to modify one of the preconfigured static values of this class
 */
 public void setRecurseFilter(VirtualFileFilter filter)
 {
 this.recurseFilter = filter;
 }
...

with this, it looks like I can implement the ejb3 deployment scanning without having to do a structural parse. This would only be needed if more control over how a deployment context classpath affects the scanning.