I worry you guys are rapidly sloshing around and churning code that was developed slowly and methodically over many years.  A lot of people have looked at an tweaked the dictionary code, but now it seems like there is a rewrite-everything festival.  When looking at potential changes, please consider that the people who came before you knew what they were doing.  If there is a pre-loop step that could have been folded into a single loop to save a little code, it isn't there because the people who wrote it are daft.  That expansion was made for a reason.

Even if you get an improved timing here or there, consider that it may vary from compiler to compiler, from os to os, and may vary widely across different kinds of applications.  Branch prediction and cache effect issues don't readily show up in simple timing loops.

Particularly as a new core dev, I recommend resisting the urge to rewrite everything you touch.  That will only destabilize Python.  In the past couple of years, two or three devs have churned an enormous amount of code.