I don't know if this will be useful to anyone, but what the heck...
This post does not propose a complete substitute for lambda, but covers most of its use cases using current python capabilities and syntax. It's a very similar idea to what Michael Hudson already has mentioned. And it doesn't apply to lambdas with more than one argument.
(There is also a module of mine, at http://www.sil-tec.gr/%7Etzot/predicates.py that implements this idea in a somehow harsh way to implement all and any predicates on iterables --check at the end of the module for example usage.)
The whole idea is a C-implemented (after all, extract et al are about speed) class, say Evaluator or Eval for short, which has an implementation for all __*__ special methods; calling (directly or indirectly) any special method of an instance of Eval returns a callable that does exactly that operation to its argument. Precedence etc are taken care of by python.
So, this way,
myf= lambda x: x**2 + 1
is equivalent to
myf= Eval()**2 + 1
attrgetter("attribute_name") becomes Eval().attribute_name
and indexgetter(index) becomes Eval()[index]
However, I have no obvious way to replace stuff as
lambda x: x[:2] + x[-2:]
lambda x: divmod(x, 6)
although in my mentioned module above I have a special .do() method for the latter case, but I presume most of you would find that ugly.
So, this idea covers the attrgetter and indexgetter use cases and then some more, but it is not a complete lambda replacement. Just my two eurocents :-|