Because brain is a highly nonlinear system, and these are hard to model. You cannot model every metabolic process in every neuron and astrocyte, you have to take computational shortcuts and use approximations. These approximations may have severe global effects.

To demonstrate the nonlinearity of these models there was an experiment http://www.pnas.org/content/105/9/3593.full where researchers excluded a single spike from the large scale brain simulation. The global state of simulation has diverged after 100 timesteps, compared to the version where this spike was present.