For more control about the sequence behavior, you may consider to simply use a custom sequence.
IDENTITY is syntactic-sugar; a feature to make a simple use case really very simple to implement.
Quite obviously there are currently bugs present in the implementation of this feature, but it doesn't mean that what the feature does needs to be extended or changed.
In many cases it is time well spent to actually consider what kind of IDs the data model should have and how those should be implemented. Personally I'd consider IDENTITY only for relatively unimportant situations - like building a test data set or something similar.
- Lars