Data blending effectively does a left join.
I recall a posting a while back that was discussing how to do a very similar data quality check and someone (almost certainly Joe) came up with a neat visualisation which helped highlight issues.
I also remember a discussion (possibly in the same thread) about constructing an artificial data source which guarantees to have all dimensions from the two datasources you wish to compare and then blending with that as the primary and the two being compared as secondaries.
Have a search on "data blending".
Edit: Searching on "data blending quality check" listed this thread first and the one from last year that I mentioned second. Here it is: