I went from UNIX in the late 1970's to mainframe zOS (MVS/OS) to VM and Linux on the mainframe. Anything you can do on an Intel box (or a room full of them), you can do on a mainframe, cheaper and more reliably, once you get past the first big financial hit. I've seen the so-called cost studies that supposedly show the room full of Intel white boxes are cheaper. Once you factor in the "unseen" costs, like the article says, and get past the startup, the mainframe looks VERY good.
Current mainframes aren't what people remember from the past. They're (physically) small, agile, and well suited to certain workloads (can you do 256 concurrent DMA transfers on an Intel box?). The problem is, the only companies that seem to be able to justify them for new workloads are ones that already have them for legacy work. IBM hasn't shown much interest in the low-end of the market (sell small boxen, then discontinue them, push licensed emulation, then kill it, etc).
Our biggest problem is finding people who know the technologies. I give classes to our Linux SA's on this, and they're usually surprised at what the current zSeries boxes can do.
Don't misunderstand, there are plenty of applications where Intel boxes make sense, I work both sides of the fence. I just hate to see mainframes maligned as "obsolete" by people who don't understand what they are now.