Will think about this one for a while.  Initial reaction is that the case isn't common enough to care about, and that the 5 byte savings isn't worth the extra code path.

> By the same logic, allowing set() as an alternative 
> to set([]) in python code is "premature optimization".

That wasn't an optimization -- it was needed for API 
consistency with other types (i.e. list(), dict(), etc).
The zero argument form supports a use case for
polymorphic creation of an empty container regardless
of type.