The power of Hadoop comes from the synergy or MR and HDFS, moving the compute close to the data. When you're talking about using Hadoop for ETL from OLTP relational tables to a DW then Hadoop will have to connect, extract data, and do the upload. Having a cluster of workers pounding the OLTP database to extract data will help little your ETL process. Even when your T phase is complex, is seldom the case that is even a blimp on the radar compared with the E of extracting from relational DB.
The more complex, IO intensive and not relational tables dependent your transformation is, the better the case for Hadoop.
Hadoop would be an obvious choice if the data would be already in HDFS. With the data located in a central RDBMS, you'll need to prove the case why Hadoop would/could help.