It's interesting to me that this article was written ~seven years ago and already feels thoroughly outdated. The core problem is that its basic premise is that we need to scratch-write an OS that is essentially the Alto WIMP (1972), a few BeOS features (1995), and questionable improvements like forcing apps through a messaging bus a la Smalltalk (1972) or CORBA (1991).
It isn't that the individual ideas are bad, per se, but if you're going to suggest write an OS from scratch then it really should be significantly different and inherently impossible to make by tweaking an existing OS, of which there are so, so many. Just an unbelievable number of OS options.