> Is there no possible optimizations by breaking LIFO dict.popitem()?

Even if there were a possible optimization, we wouldn't care.  The API is too desirable to forgo in the name of micro-optimization.  We don't design our APIs that way -- trading the relevant and actionable for ethereal and unknown.

Also, it is common sense that addition of a key/value pair at the end is a readily undoable operation (no more expensive to remove than it was to add). There will be a dummy entry in the hash array just like there is now.