Via oDesk developers I've been conducting a series of comparative tests between the two main Open Source ETL engines: Talend Open Studio and Pentaho Kettle.

Pentaho

Pros

• Wide standard inbuilt function set
• I believe it has Open Source web-based processing (can run jobs as web-service)
• Also contains the Weka data mining tool

Cons

• Pentaho website is very thin, little content

Talend

Pros

• Has a more "rich" embedded Java component that breaks processing down to setup, per-row, teardown
• Appear to be more downloadable components, well organised in TalendForge
• Lots of good quality training
• Eclipse based, nice and familiar
• Coherent, tidy, thorough UI (makes Kettle appear a little amateurish)
• tMemorize allows easy access to previous rows, without custom Java code

Cons

• Appears to have a more limited function set
• Customer engagement feels heavily focussed on up-sell

Maybe

• Generates Java before running Job. This may be a 'con', due to fragility or performance. It may be a 'pro' due to having transferrable Java jobs.