Do we know how often this is blocking on a subframe beforeunload vs a main frame beforeunload? It's possible that  issue 692224  will mitigate this somewhat (by simply not running them as much)

I don't know of any tracking of whether it's subframes or parent pages. As per the latest discussion in  issue 692224 , it won't help here because we're still running the code. We're just not allowing for showing the dialog.

Issue 365039 should help with this, where we won't try to run the beforeunload handler if there isn't one registered.

As I mentioned on  issue 701434 , I'd be concerned about slow machines not having enough time to tell the user that they're going to lose data if we lower the timeout too much.  Maybe some UMA stats could give us a sense for how often this happens?