Testing indicates that multithreading layered table drain can result in speedups of 2.5-3x in specific scenarios, for this reason I am recommending we implement a simple multithreading solution to enable faster step up times in WiredTiger.

From my testing around 8 threads in a 16 cpu machine was the magic number, so initially we will hardcode the number of threads to be 8, or half the CPU count if possible.

To multithread we will:

• Create a queue of work items (ingest table tree IDs)
• Spawn a thread group, which will drain the queue and run the relevant ingest drain steps

This is the simplest version of multithreading, effectively delegating a tree to a thread. There are other implementations possible but testing doesn't necessarily support the increase in complexity yet.