Possible memory leak in 3.0 (same workload, regression from 2.5.1)

Steps to reproduce:
-Create 4-node cluster (3.0-597)
-Load beer-sample
-Run this load-generator from 2 separate clients (same command):
cbc pillowfight --host <IP> -C HTTP -b beer-sample --num-threads 4 --min-size 1024 --max-size 1024 --ratio 50 -Q 4 -I 10000 --loop
-Observe memory usage constantly increasing. On 2.5.1, it stays flat at ~30mb per node. On 3.0, it is already up to >700mb per node after an hour

Let me know which stats are most useful to grab