The SQL Server tuning advisor has historically been terrible for any database of any complexity at all. It imposes an enormous load, and the results are usually contradictory and often very sub-optimal if not detrimental. Works great for a hello world style database, though.

There was a post earlier (https://news.ycombinator.com/item?id=19422554) that I assume inspired this post and it was someone who benefited by correcting the DBMS' very wrong assumptions about the cost of random IO, and this was always the Achilles heel of SQL Server as well with any sort of fast storage -- it absolutely refused to accept that anything could be faster than about 150 IOPS, and would try to brute force through with terrible approaches. In this case the guy could just change the configuration and it would properly calculate query plan costs and execute accordingly, where SQL Server, to my knowledge, still doesn't let you change this costing, nor does it dynamically determine it.