3.5x throughput drop in recent runs of 'KV : DCP, 250M x 1KB items, DGM' perf test when using new DCP test client

When checking the daily showfast perf results for "KV : DCP, 250M x 1KB items, DGM", I noticed that recent runs (from the last 2-3 days) are showing a ~3.5x reduction in throughput - see http://showfast.sc.couchbase.com/daily/#/history/KV%7CDCP,%20250M%20x%201KB%20items,%20DGM%7CAvg%20Throughput%20(items/sec) :

Both builds 6.6.5-10072 (highlighted) and 7.1.0-2020 show a drop in throughput; these were both run in the last few days:

• 7.1.0-2020: http://perf.jenkins.couchbase.com/job/triton/43970/ , Started 2 days 16 hr ago, Took 1 hr 20 min on triton-client
• 6.6.5-10072: http://perf.jenkins.couchbase.com/job/triton/43984/, Started 3 days 11 hr ago, Took 1 hr 47 min on triton-client

There hasn't been many changes in KV-Engine on mad-hatter branch (only backports), so this is either an issue with a recent backport or someting environmental.

Starting initial binary chopping on kv-engine builds (are fewer changes on mad-hatter so starting there).

mad-hatter Changelog: http://changelog.build.couchbase.com/?product=couchbase-server&fromVersion=6.6.4&fromBuild=9918&toVersion=6.6.5&toBuild=10072&f_asterixdb=off&f_backup=off&f_cbas-core=off&f_cbft=off&f_cbgt=off&f_couchbase-cli=off&f_couchdb=off&f_eventing=off&f_go-couchbase=off&f_go_json=off&f_goutils=off&f_goxdcr=off&f_indexing=off&f_kv_engine=on&f_n1fty=off&f_ns_server=off&f_query=off&f_testrunner=off&f_voltron=off