Some[1] have argued that unicode is better in 2.7 than 3.X. Async looks very useful, though, so I'll add that to the list including the subTest context manager.
High quality unicode? Where every string with even a single emoji takes 4 times the memory. They should have used UTF-8 for their internal representation.
Putting that aside, my issue with 3.X is that I have to use the unicode type even when I'm manipulating byte strings.
The bytes object is not a string type. It's treated more like a list of integers. It has very little string operations. Only in 3.5 did they add back the ability to % format.
Not to mention all the b prefixes it just becomes painful.
http://lucumr.pocoo.org/2014/1/5/unicode-in-2-and-3/