Results show that the size of replicated data is high with attribute-based granularity.

The average replicated (and response) sized are
((1.7681421793709815E7 + 1.755308562069113E7) / 5 743.8) * bytes = 5.99058091 kilobytes
whereas this cost should be minimal as only an int is being replicated.

https://hudson.qa.jboss.com/hudson/job/eap-6x-stress-http-repl-sync-granular/4/console-perf17/