Hmm, I think this could confuse people into thinking that the queries displayed were in fact prepared queries.
Maybe we could gather some more ideas.
I think thats a reasonable concern - the main benefit of $1 is that its already designated as something that can replace a constant, and still be read by the Postgres parser.
Is there any other character that has the same properties?
I'll also note that Greg Stark mentioned in [0] that "There's another feature pg_stat_statements *really* needs. A way to convert a jumbled statement into one that can be prepared easily. The use of ? instead of :1 :2 etc makes this a mechanical but annoying process."
Using $1, $2, etc. for jumbling statements would give us that for free, no additional effort needed.