10x slowdown in K/V pillowfight perf compared with 3.0.3

Comparing the performance of a 3.0.3 cluster to a 4.0.0-1891 cluster, I see a 10x drop in front-end op/s driven using pillowfight - from ~25k to ~2.5k - see attached screenshots.

(I'm very surprised by this, and assuming / hoping I've messed something up in my tests, but nethertheless this is what I'm seeing...)

Steps to reproduce:

• 2 node cluster, each with 2 CPUs, 4GB RAM.
• default bucket quota of 100MB, 1 replica.
• Use pillowfight with 1000 item working set, doc size set to 100bytes:

$ cbc-pillowfight -U couchbase://192.168.46.101/default -m 100 -M 100