There are a lot of reasons to favor relatively small objects with crystal clear behavior and responsibility. The theory stretches back to the 70s with information hiding, low coupling and high cohesion. The practice pays off every day.
Larger objects that gather up more responsibility, behavior and data become harder and harder to develop,
test, modify, reuse, read, deploy, etc.
There are some
patterns like Facade that put large grained interfaces in front of small grained objects to protect clients of a service from having to understand all the tricky bits within.
Fine grained objects often appear in rich domain models with lots of interesting behavior and well hidden data. In architectures like stateless web servers you can wind up with an
Anemic Domain Model that looks good this way at first glance, but turns out to do nothing but push data around.
Any of that help?