The issue is what tools you have to dig in when you're asked, "Why is this page taking 5 seconds to load?" With a traditional single threaded application there is the inherent simplicity that you can profile it, look at timings, and see where the performance went. With an asynchronous distributed application, you have to do a lot more work to start digging in.
The reason why this matters is that there are always some boneheaded performance mistakes. They would be trivial to fix if you only knew what to change. Without visibility, you won't be able to find where they are - you're just stuck suffering the consequences.
I'm definitely not saying that this is impossible. Far from that - Google succeeds brilliantly. But the kind of behind the scenes pervasive visibility that you need is an essential component, and it is not something that happens by accident or is trivially retrofitted on.