I've noticed that Table.ML seems fairly slow when it's the insertion of 1/2 elements into an empty set. Looking at the code for Leafs it seems it does a unmake goes through a bunch of code for branches before it goes back to a make call.
If you believe one can get a performance improvement that is noticeable in practical Isabelle applications, I suggest that you make the the corresponding change and compare the build times of, say, session HOL-Analysis with and without the change.
Last updated: Feb 27 2026 at 20:31 UTC