It will be interesting to see how Guido et al will judge the py3k transition-strategy in hindsight and whether they will repeat it or move back to a more traditional, incremental development model.
"Oh wait, this machine has foo 1.5.2, not 1.5.6!!!"
Though a solution to this scenario is to avoid including too many libraries in the standard distribution in the first place, so that incompatible changes in those libraries don't affect the base.
This in turn means that big packages with lots of dependencies would need to say "you need fooliba-1.5.6, foolibb-4.3.2, etc" instead of just "you need foo-5.2", but other languages do this and they seem to manage it alright.
I'm not saying one or the other is definitely better, just that I'm looking forward to the final judgement after this multi-year effort.