Thanks for the followup. It sounds like the optimizer isn't doing what it should. Most scalar subqueries can't be rewritten as a single query with a simple join. Typically, as in the link you provided, they allow you to combine different levels of granularity: averages combined with non-average data, for example. When the queries can be rewritten, the execution plan is usually the same (and therefore the execution time). I'm guessing that the statistics in the database are wrong in this case. Those statistics are updated by analyzing each table. No big deal. As long as things are working for you, I'm sure you're not too worried about it.
I'm an Oracle DBA by day, so this sort of thing bothers me.