[FTS, scorch] high memory consumption by cbft when there are items in persistence queue

My setup:

node211: data service, 256GB RAM
node39: FTS service, 10GB RAM, FTS quota  is 1GB 
node40: FTS service, 10GB RAM, FTS quota is 1GB

10M documents, wiki dataset
 FTS index size is 14GB, so 7GB per node

Test:

"wild card" queries and docs mutations (mutation rate is 1K docs/sec)

Behavior:

Phase 1 - indexing. 
While indexing memory consumption by cbft is up to 3.5GB
Index is built w/o any issues

Phase 2 - warm-up
During warm-up phase client runs "wild card" queries only (no kv mutations)
The memory consumptions remains on 3.5-3.6 GB.
Consistent query throughput, no issues

Phase 3 - perf test

Once the amount of items in "Items to persist" starts to grow the cbft memory consumption quickly goes up. With just about 500 items in the queue cbft allocates another 2GB of RAM
With 1K+ items it takes all available memory (in my setup) and node gets failed-over due to OOM issues 

Validating other queries and memory setups...

Logs from one of the runs:

https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-atlas-sdk-enduser-469/172.23.99.211.zip https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-atlas-sdk-enduser-469/172.23.99.39.zip https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-atlas-sdk-enduser-469/172.23.99.40.zip