The assertion that microkernels are slow makes me think you're out of touch with technology for 1-2 decades? L4 is not that new anymore.
On the other hand, promoting a JVM and garbage collection in hardware means you can't be further back than maybe 4 decades, since this is when they did the exact same thing and called it a Lisp Machine.
Of course, writing in Lisp requires quite literally PhD-level understanding of algorithms and software design, just like a microkernel. So in the end, maybe it always was the better solution to just use a scheme that even a monkey could use, and then make a lot of monkeys all excited about it. Divide and Rule^WConquer.
But yes, I think that AT conclusions are a bit strange..