That's what I did after a nice stint at Microsoft, and it's amazing --- working on compilers is even _more_ fun than you'd think!
But, there's always the gloom & doom at the end. Post-doc positions are easy to find, but there are very, very few academic research+teaching positions and competition is quite intense. Still, it's a great ride so far. And the worst-case "build compilers for a big bank" scenario isn't so terrible.
I feel that there are certain aspects of cs that I really want to focus on for my own personal benefit. Sure I could study on my own (and do), but being in a program of pure laser focus for 3 to 5 years would be deeply rewarding for me.
I would most likely study some fusion of cs and art. These days I don't feel like I have enough time to move fast enough. I am trying to get into gpgpu (via CUDA at the moment) and I need to get into hardware. Having the full time to do this, without needing to worry about financials with my current lifestyle, is my dream.
I did a PhD (not in CS) because I wanted to understand a set of scientific problems in depth, do the research, understand setting up experiments, monitoring them, etc. There are other ways, but it takes 2-3 years of semi-structured work to wrap your head around some of these aspects. At the same time, it helps to understand what you like/don't like and try out new methodologies and areas of research.
Not the only way to get there, but a very fun one.
