Changing the database model on a live solution in quite unpredictable.
1 - it's unclear why creating an unstored calc requires to process records (at least that's why the non-progressing progress bar indicates)
2 - sometimes, a change is applied in no time, sometimes it takes ages, why? remains a mystery.
3 - applying a change on a large table almost stalls the server. When you look at the CPU usage, you see that only 1 is working (whether it's due to a database change or to an "implicit" indexing caused by a find.
I have no idea how feasible this is, but making those more predictable and multi-threaded would ease our daily work in many situations.