An annoying, but useful, hurdle
Remember the 640K limits of DOS and how they plagued PCs well after systems were available with MBytes of RAM? Or how PRINTs to Windows consoles are still very slow in the age of super-fast GPUs? Remember wishing they would just fix it and damn 10 year old codelines rather than forcing everyone to live with bad decisions forever?
I don't think Python has any such obviously dim features built into it, but over the 15 or so years it has been around, many little quirks have accumulated and are a hassle. Python 3.x is meant as a clean up to facilitate working with the language for the next decade(s).
For example, I've always thought Python is great to discover new problem domains. But Python 2.x and Unicode is not exactly ideal and I still don't grok Unicode at low levels. If 3.0 really brings easy Unicode, I am well willing to deal with upgrade hassles.
Google "Let’s talk about Python 3.0" and you'll see a much better argued, and quite entertaining, version of my argument. Love the banana and monkeys story!
I am pretty much each change has been discussed extensively to weigh the benefit vs. the cost and the designers are _extremely_ anal about keeping the language as easy as possible to understand. I trust them. And if they were wrong, Python 2.6+ will be around for a long time.
Besides the regression testing workload, I think the biggest risk is when you depend on multiple non-core 3rd party libraries that move, or do not move, to 3.+. And, for the folks concerned about the 10% penalty, just relax and wait a more-optimized Python 3.1, it'll come. Get it right, then speed it up. Basic Knuth.
Even M$ damned the torpedoes and broke some compatibility with Vista. Oh, wait, that's not exactly proving my point, is it?
p.s. This story's title is just a wee bit sensationalist and alarmist, IMHO. But txs for the publicity nevertheless.