Counterargument: you don't always know when you start writing a document how many levels of heading you're going to end up needing. A rough outline is inherently top-down, not bottom-up, so keeping your granularity options open makes sense.
Not sure I follow. You're not going to suddenly discover after two days of writing that you need a top-level header for the whole document, but you might very well discover that one of your "leaf" sections needs to be broken up further. The difference being that in your scheme I'd need to change all existing headings in the document to handle that.
Another (minor) thought: if you're the type who starts by outlining the whole doc with headers only, the Markdown approach naturally makes indentation follow structure.
Meaning that if you start off by writing all your top level headings like a normal human being, you'll have to go back and change every single heading once you realise you need a h6?
Atleast in most common use, markdown parsers aren't that smart. They don't do much lookahead, they parse text in a stream with little buffering.
The most complex behaviour is tables, and even there you can easily spot where the parsers simply go character by character along the line with no difference other than being in "table mode" (or for code fences, in "code mode")
