Unfortunatly I can’t share the data, but you can try it out on any data set you have
change the ontology namespace to “/#” by adding an extra “/” before the “#”
You will see during inference, there is a lot of delay compared to when there is no extra “/” before the “#”.
There is nothing in reasoning that is sensitive to the characters inside IRIs. Different IRIs would change hashcodes and iteration order for some internal operations but it is unlikely to have a significant difference due to that.
Is it possible that making the change in the IRIs changes how definitions in your ontology interact with each other or with the instances? If you can share the query plans for the slow and fast versions that would help us to understand what might be wrong.