Lars/Martin,
As I read somewhere, Delta store has two parts. L1 and L2. L1 is Row Store and L2 is Column Store (but with no compression). And the UPDATE process is also a two step process. First records are soft-deleted in the Main store and then updated records are INSERTed in Delta store which will eventually be merged into the Main store.
This means that Hana first has to uncompress and materialize all records (with all columns in the table that is being updated) in the Delta store. But isnt this a very unoptimized way of UPDATing data? In a data warehouse such updates are frequent. And in case you have multiple updates, every update query creates a separate copy of the data in Delta store.
Also, with Columnar, shouldnt the DBMS be intelligent enough to understand the specific columns being updated and pull in only those fields and not all the fields into delta?