By adding a method in the CONFIG that skips that, for large tables freeing the table itself if it is large (65/128kb InternalTable) is rather slow (0.2ms+ - imagine having to release 1000's of those in a STW pause), probably a problem with malloc with large(r) blocks of memory though. I think table sizing heuristics could be tuned to keep the table size anyway.
From current POV that will probably workarounded by moving deletion of the CHT into the concurrent phase.