Haven’t many similar issues been solved in journaled file systems and/or things like database transaction logs and indexes? Real-time high precision accuracy is not required, knowing how big a directory is, is a frequent use case of directories. Hell, ‘df’ tracks this at the partition level, including edge cases, as does ‘du’
(Both databases use MVCC (https://en.wikipedia.org/wiki/Multiversion_concurrency_contr...) to ensure that concurrent queries all see the database in a consistent state. That makes it necessary to visit each row and check their time stamp when counting rows)
