Besides actual mistakes and inconsistencies noted in comments below, in general for me this change adds even more confusion to this 'runtime management' area of the code...

From what I understand, the issue here is that BKE_mesh_runtime_clear_cache (unexpectedly) also clears the mutexes? Then I would simply add a new BKE_mesh_runtime_clear superseding function, that does clear mutexes and then call BKE_mesh_runtime_clear_cache, that clears everything in runtime besides the mutexes. BKE_mesh_runtime_clear would be used from mesh_free_data, while BKE_mesh_runtime_clear_cache would be used from BKE_mesh_clear_geometry.

Then having (static, private) helpers to factorize out mutexes handling would be good indeed, although I would only have create and free ones tbh, the reset one does not seem really necessary here?

This would also need comments to properly document what is expected behaviors/roles of all those functions.

You're right there were some bugs in the patch.

After trying some different approaches I moved the clearing of the runtime struct to the read_file of mesh and use init_data/free_data as higher level API.