Layered cursors have a simple algorithm for when they reopen the stable table - they'll do that whenever the "external state" has changed. Meaning a step up or step down or picking up a checkpoint. Consider a workload of 1000 collections - 100 being actively written and 900 not being written but being read. Sessions on the follower are using all 1000. When a checkpoint happens a session will need to reopen the stable table for all 1000 of them (instead of just 100). Now multiply by several hundred sessions, and what should be an (expected) smaller bump at pick up time becomes a much larger bump.