WiredTiger checkpoints work best when there is not too much dirty data in cache. In that case, the existing "write leaves" phase isn't necessary.

At the other extreme, in very high throughput update workloads, the write leaves phase is unhelpful because the cache is still 100% dirty when it completes, so all it does is slows checkpoints down.

A better approach would be to allow eviction_dirty_trigger to be set automatically by the system to bound the time taken for checkpoints. For example something like this:

wiredtiger_open(..., "eviction_dirty_target=20,eviction=(checkpoint_target=10)");

would allow the cache to become 20% dirty but would scale that down so that checkpoints take no more than 10 seconds.