The reality however is that many applications find a recursive descent parser easier to handle once the grammar has been fixed (and in spite of its many caveats

The reality however is that many applications find a recursive descent parser easier to handle once the grammar has been fixed (and in spite of its many caveats!), so PEG is a natural formalism for them. For example I once tried to write a full parser for Lua in LALRPOP and gave up after a week because it was too slow to compile and too hard to avoid conflicts; I ended up writing a recursive descent parser in a single day. I already knew that LALRPOP doesn't resolve shift-reduce conflicts and also how to transform the grammar to avoid them, but that wasn't enough for me. I think there is a niche for CFG-like formalisms with bounded PEG-like constructs so that it can be quickly adjusted without giving up all the benefits of CFG. I have been already in situations where I really want to have such formalisms for multiple times, including the aforementioned one.