WT cursors are cached by both WT and the server. On Mongo's end they are cached by tableId (code pointer). We may be able to improve the performance of secondary oplog application of ci/cd entries if we use this cache effectively. During an index build, container ops will generally all involve the same handful of containers.

Some naive options are to derive a stable table id from the container ident, cache by ident directly, or perhaps bypass the Mongo WT cursor cache and leave cursor caching entirely to WT. Weighing these options will likely require a functioning primary index build to benchmark against.