this primarily affects CASSANDRA-1314 and to a lesser degree CASSANDRA-981
In combination with SimpleSnitch from https://issues.apache.org/jira/browse/CASSANDRA-1314 this patch allows for much better cache utilization resulting in a full order of magnitude decrease in disk reads across my entire cluster.
as a side benefit, it was simplest to refactor things so that local and remote reads in a multiget are done in parallel now (before, we would do all local reads, then all remote ones).
Integrated in Cassandra #503 (See http://hudson.zones.apache.org/hudson/job/Cassandra/503/)
update client_only example for CASSANDRA-1317 changes
Patch by eevans