LruTaxonomyWriterCache uses LRUType.LRU_HASHED by default. This has a very small but non-zero chance of producing incorrect results due to collisions in the longHashCode of FacetLabel. If such a collision occurs, then an affected document will get an incorrect facet.
This has happened to us in production. While it is a rare occurrence, the consequences could be significant, and it was not immediately obvious what caused the problem. Therefore I think it is better if the default is changed to LRUType.LRU_STRING, which is guaranteed to be correct.
I will add a patch containing this change as well as a test for the issue.