Looks like a good start.

I think you should replace all of the lines:
 if isinstance(other, dict):
with
 if isinstance(self, type(other)):


Since if other is an instance of a dict subclass, he should be the one to process the addition. On the other hand, if self is an instance of the derived type, then we are free to do the combination.


I think you should also change this wording:


"the result type will be the type of the left operand"


since the result type will be negotiated between the operands (even in your implemenation).


__sub__ can be implemented more simply as a dict comprehension.


Don't forget to return self in __isub__ and __iadd__ or they won't work.


I think __isub__ would be simpler like this:

def __isub__(self, it):
 if it is self:
  self.clear()
 else:
  for value in it:
 del self[value]
 return self

I don't see why you would bother looking for keys (iter will do that anyway).