access scanner took much time to run

Created this bug to track the performance of access scanner.

Some facts from vms:

Rebalance test with NRU change. Here are some quantified facts: (all timestamps are based on UTC)

> access scanner scheduled at 2, finished at 05:53:40, took almost 4 hours for 10M items.

> rebalance started after an hour, at 06:53:41. During that time, the cluster were dealing with mixed traffic to build the working set.

> rebalance finished at 07:39:12, took 46 minutes (matched with 2731.88 seconds on the pdf)

> rebalance avg ops was around 2k (eyeball)

Is it expected, or something needs improvement?