Well, it was just installed a few weeks ago - hasn't seen a whole lot of use yet, so not too many experiences on which to base an opinion.
A single run costs ~$400 (an order of magnitude less than an Illumina Hi-Seq run!), but only gets you 10-20 million bases of sequence. (A human genome has ~3 billion bases, and you need ~10-fold coverage to make sure you get each base. In the Nature paper, they sequenced Gordon Moore's genome - and it took about 1000 chips to do it.) There's some talk of using it to QA libraries before a Hi-Seq run - for $400, I think that makes a whole lot of sense.
I'm not sure that talking about Illumina approaching the magical $1000 genome is quite on target, either. The other thing most people don't talk about is read length - an Illumina run will get you enough sequence, but it's only in 50-100 base chunks. Reassembling an entire 3 billion base genome (6 billion if you consider both haplotypes!) is either extremely difficult, or actually theoretically impossible, depending on who you ask.
The $1000 genome X-prize is to sequence a single genome, for $1000, in one week, to the quality standards of the original Human Genome Project (1 error in 1000 bases, IIRC). To do that, you're going to need long-range sequence data. I think that the nanopore sequencing approaches, that read a single piece of DNA for hundreds of thousands of bases at a stretch, have a much better shot at getting there in the next few years.
The other interesting thing about the Ion Torrent PGM is that the magic happens on the disposable chip - and that can be upgraded without the rest of the machine (sensor interface, fluidics) changing. The next generation of chips is spec'ed to produce 10x as much sequence, and it's already sampling at early access laboratories like Baylor and MIT.