Slow down in disk persistance when updating items in heavy DGM

When updating items in heavy DGM (40% resident), the disk drain is slower than draining of new items.

Ex:
500 items per second for new items, but only 300 for updates.

Loading 20K items into a cluster of 2 nodes at 250 ops with 1 replica.

We loaded 500K items, pushing the resident ratio down to 40%, then updated the same 500K items, during this update phase the disk drain rate dropped down to 60% of what it was before.