Failover not correct for clustered webapp

To reproduce:

• Run httpd (downloaded from mod-cluster package)
• Copy 'all' into node1
• Modify node1: add listener, jvmRoute, depends on HAModClusterService (in jboss-beans.xml)
• jvmRoute="${jboss.jvmRoute:nodeX}"
• Copy node1 into node2
• Start httpd (httpd.conf attached) on 192.168.1.5:8000
• eth1 = 192.168.1.5
• Create IP alias for node2 (as root): ifconfig eth1:1 192.168.1.9
• Start node1: ./run.sh -b 192.168.1.5 -c node1 -Djboss.jvmRoute=node1
• Start node2: ./run.sh -b 192.168.1.9 -b node2 -Djboss.jvmRoute=node2
• telnet 192.168.1.5 8000<cr>info<cr> shows node1 and node2
• Depoy web.war (attached) to both node1/deploy and node2/deploy
• Open http://192.168.1.5:8000/web/setup.jsp (creates state in the session)
• Open http://192.168.1.5:8000/web, reload a couple of times
• Let's say the session is created on node2 (jsessionid should have "*.node2" as suffix): kill node2 (CTRL-C)
• When node2 is down: reload the webapp
  ==> An empty web page is returned, or 404 is returned
• telnet 192.168.1.5 8000 <cr>info<cr> shows that node2 is gone (BTW: this takes 30 secs, why isn't it immediately removed like a context ?)
• The jmx-console on 192.168.1.5:8080 shows for the session cache that the contents were indeed replicated !