Good support for strings that you want to treat as text but have unknown encoding, the biggest example of which is all OS strings on posix systems. Being forced to treat all strings that aren't guarenteed to be valid unicode as a series of integers makes everything harder - parsing text-based protocols, text-based file formats, hell last I checked it was very non-trivial to write a version of "ls" which didn't crash if the directory contained a file called "\xff".

Massive compatibility with the biggest installed Python codebases?

People already know it?

The most used?

All points I listed above.

And still hugely important.

ekimekim said "even without considering dependencies/migration cost". Let's keep things in context.