I feel like you stopped reading after the first sentence as we ended up saying the same thing. I certainly don't mean to say that determinism implies predictability in the first sense you define above. In the second, more general sense "knowing a future event before it takes place" predictability really could depend on the concept of emergence (or solipsists notion of macro-levels). Here's a thought experiment: It's conceivable that you could perfectly simulate a higher level of organization in a system (which would require less information). It's also conceivable that objects on that level (emergent properties of the lower levels) operate according to their own rules distinct from (but derived from) the lower levels. Here's the kicker: given that the higher level is an emergent property of the behavior of the lower level, when we look at it in isolation, it contains much less information than the lower level. In fact several state iterations of the lower level could pass before a state change percolates to the higer level. Thus, in such a thought experiment by having a perfect model of the higher level, you could outpace the lower level and know a future even before it takes place.
An additional (and perhaps stronger) argument is that Rice's theorem states that you cannot in general predict the behavior of deterministic computation. You can't know if a program will ever access the network card, output a letter "A," halt, etc. even though all of these programs are known to be deterministic.
I don't think anyone is saying that we can create a cellular automata (or any computer) capable of simulating the entire universe. That doesn't mean that the universe can't be modeled as one. You can't simulate an Intel i5 CPU in realtime in Windows 7 running on an Intel i5, but that doesn't mean an i5 isn't deterministic.
As a side node, the potential richness and unpredictability of deterministic computations is a major theme of many works by Rudy Rucker. Any investigation of cellular automata will reveal the unpredictability of deterministic computation, which is essentially the whole point behind cellular automata.