Sorry, I will claim ignorance on this one. I don't recall the last time
I've used a bisection, but it was probably around the time bisect.py was
first added to the standard library.  I do recall using heap sort as a
way to compute the top N items.  I've always done that by explicitly
mapping the raw data to a list of (key, value) tuples.

The arguments worrying about bad design make some sense to me;
consistency isn't always the panacea that people want it to be...