XDCR checkpointing: Post destination bucket flush/recreate, replication very slow, seen as periodical spurts

The most recent fix to fetch remote vbopaque and replicate after 60s of idle xdcr works but with the following observations -

1. Doc replication is in spurts, with mutation replication rate <500/sec, replicators under 3-4% utilization, doc checks rate spikey(falling to 0 regularly)

2. Outbound mutations = 0, percent completed = 100, active vb reps =0, waiting vbreps =0 throughout this replication

3. In all replicating 20K items takes a little more than 5-6 minutes. Changing checkpoint interval from 60s to 1800s made no difference.

4. The first spike after 60s is rather "lonely". The next spike comes after 10s and rest come in 2-3 sec interval. (see screenshot)

If we do a bulk vbopaque fetch, why aren't we seeing steady replication?