You're absolutely correct when you say...

When I teach T-SQL methods that do require some formatting for the output, I tell folks that they still need to keep the Data and Presentation layers separate, especially since formatting can be such a load on the code.  Just like you say above, do the heavy lifting and aggregations first, which also usually greatly reduces the amount of data that has to go through any type of formatting including even something as simple as pivoting the data.  Although both a CROSS TAB or the PIVOT operator requires aggregation, it's a rather expensive aggregation because of the decision required to determine which column things must go into.  With that, I do what Peter "PESO" Larsson calls "pre-aggregation", which first aggregates the data to the smaller result set and then do the CROSS TAB or PIVOT on that greatly reduced data. 

On the continued used of FORMAT... I'll never use it so long as it has such terrible performance even if the code is more clear.  My feeling on all of that can be summed up by two points...
1.  I won't use something that's got a built in performance problem because performance and resource usage are second only to accuracy and it's a very close second.
2.  If someone doesn't know how to use CONVERT, they should do a bit of studying.  It's a part of their job to do so.  It's like saying that I shouldn't use a Tally Table (or equivalent function) and should use While Loops instead simply because someone might not understand what it does.  A well placed comment in the code can also help there, as well.