In terms of performance, the primary rationale for only retrieving columns you need would be to allow the database engine to select a proper index to use, specifically, one that covers all the columns you are making use of. This can actually have a massive impact on the execution plan, far more than simply fetching say 2x more data, as you point out.
In fact, I'd go so far as to say that the practice of selecting all columns makes it practically impossible to make use of covering indexes for high impact queries.
In fact, I'd go so far as to say that the practice of selecting all columns makes it practically impossible to make use of covering indexes for high impact queries.