Hi szawadski,
The examples helped alot to understand your problem.
yes I think if you do not want case (2) to be updated then it will get hard to to with the current knime db nodes.
I think in this case you will either need to first use the DB Query Reader to read all existing rows to KNIME before the merge (to filter out the existing rows without updates)
Or write the update contents into a volatile/temporary or stating table on the database -> then do the merge with an DB SQL Executor
I guess you could also do it with an ugly solution within KNIME… as the current Merge node does not support existing rows without changes but I do not think that would be better then using a volatile/temporary tables
~well one solution in knime could be the following for (2) while doing all the other cases with the normal merge node:
Use the DB Table Selector and Parameterized DB Query Reader to select all rows in your merge data with all equal columns (or the columns you want in (2)
Then use the refernce row filter to filter these cases out before your merge
Then do the merge
*and thw Parameterized DB Query Reader would just use the SQL you want for checking (2) 
