According to Norman, there's some important memory optimisations coming in this version, that should in theory allow us to get rid of the custom replaying decoder.

I'll do the upgrade in a separate branch and I'll pass on to QE to run:

• Data stress tests and compare with current version
• Client stress tests and compare with native Memcached and previous versions

Depending on the results of this, we can potentially consider doing the upgrade in 5.1.x.