Efficiency of object generation

Random question regarding optimization: for relatively small objects (five
or six attributes) that need updating, is it more efficient in Moose to
generate a new object, or to access an attribute and update it?

NOTE: in order to avoid errors in my program, the existence of the object
must be checked in a (possibly large) hash before updating, whereas
generation can be done spontaneously.

I know, I know, this question is fairly vague. I'm happy to explain the
context of my problem, if that helps.

Thanks,
Faelin