The field equations of general relativity can be separated into
hyperbolic evolution equations and elliptic constraints [ADM].
The evolution equations propagate an initial field configuration
'forward' in time, similar to other theories of classical fields.
However, due to the constraints one cannot choose the initial
configuration freely and this is very different from other classical
fields. In some sense the constraints 'connect' spacelike points and
thus one could call general relativity 'holistic' if this would
not be such an abused word.
We don't really know what the quantum theory of gravitation is,
but one would assume that the classical theory reflects the properties
of the underlying quantum theory and indeed the Wheeler-deWitt equation
is nothing but the operator version of one of the constraints.
I think one needs to keep this in mind when discussing thermodynamics
of general relativity, the information loss problem or the entropy of
black holes. E.g. if one specifies the metric near the horizon of a
(near spherically symmetric) black hole, the constraints already
determine the 3-geometry; Therefore I do not find it surprising that
counting microstates provides for a holographic result which differs
substantially from the naive expectation.
I would also think that an approach to the information
loss problem which emphasizes locality as 'conservative' is misguided.