Some Work instances would benefit from being queued according to a priority ordering.
An example would be fulltext indexing, which we'd want single-threaded but where the "quick" metadata changes should go before the "slower" fulltext extraction from binaries.