[FTS] 4x performance degradation after net and text libs update (multi-node)

There is no regression on single node setup.

As usual for multi-node issues the most affected are light-weight queries.

The "net" and "text" updates have been introduced in build 2530. Below is the comparison of low term throughput on builds 2528 and 2530. There are two issues actually: the throughput on 2530 is low. But it gets even slower at some point (and latency increases).

Investigation is ongoing.

Stats, profiles and logs:

build 2530 - http://perf.jenkins.couchbase.com/view/FTS/job/atlas-sdk/1669/

build 2528 - http://perf.jenkins.couchbase.com/view/FTS/job/atlas-sdk/1668/