Null keys are now bypassed for uniqueness checks. (V. Horsun)

If a new key is inserted into a unique index, the engine skips all NULL keys before starting to check for key duplication. It means a performance benefit as, from v.1.5 on, NULLs have not been considered as duplicates.

NULLs are ignored during the index scan, when it makes sense to ignore them. (A. Brinkman).

Prevously, NULL keys were always scanned for all predicates. Starting with v.2.0, NULL keys are usually skipped before the scan begins, thus allowing faster index scans.