Every create and drop causes an entry to be added to the shared metadata queue (conn->disaggregated_storage.shared_metadata_qh).  On the leader, these entries are resolved and removed at checkpoints.  But on the follower, where the create/drops are applied via oplog entries, the entries are never removed.  So if 10K tables are created over 100 checkpoints, we have 10K entries on the queue.  And that queue is walked during checkpoint pickup for each new table, leading to extra expense.  Obviously if there is a lot of table churn, this will get very expensive over time.

This ticket is related to WT-18310, however these tickets have different outcomes (correctness for that vs perf for this one), and probably different fixes.  This one is noticed on pickup, that one on step down.  And we have a reproducer for this one.

Reproducer is using the latest version of bench/dhandle.