Memory growing to over high water mark and breaking down cluster as intra-replication queue grows

See logs at:
http://s3.amazonaws.com/customers.couchbase.com/mb_11734/collectinfo-2014-07-15T163255-ns_1%4010.196.89.226.zip
http://s3.amazonaws.com/customers.couchbase.com/mb_11734/collectinfo-2014-07-15T163255-ns_1%4010.198.23.223.zip
http://s3.amazonaws.com/customers.couchbase.com/mb_11734/collectinfo-2014-07-15T163255-ns_1%4010.198.37.111.zip
http://s3.amazonaws.com/customers.couchbase.com/mb_11734/collectinfo-2014-07-15T163255-ns_1%4010.198.52.92.zip

Same workload as mb-11643. With XDCR and views enabled, the intra-replication queue grows uncontrollably as does the mem_used (with no new items being added, only updates). Eventually it reaches over the high watermark, ejections begin, the memory goes down and the cycle repeats itself with the memory growing again.

See the last 5-10 minutes of logs as there was some unrelated activity beforehand