Closing as WNF. Future improvements in this area will be dealt with as enhancements, and leyden AOT caches should help cache generated MH adapters.
22-10-2025
I'll take a look. I'll note that the absolute regression is 25-30ms, and very much in line with the overheads we've measured when bootstrapping some MH-powered reflection on targeted startup micros. This might be a good place to deal with any startup improvements we can come up with, though.
22-11-2021
[~redestad] can you help look into this startup regression?