If you don't make that assumption, then people aren't responsible for their actions and there wouldn't be a basis for law revenge.
FTFY. We would still find it necessary to lock up criminals for the protection of society.
Indeed. Also, you can make the case that the incarceration of criminals should be carried through upon because of its deterrence effect on other people, who would otherwise have (through such deterministic patterns) carried through on this actions not because of their own will, but because of the absence of punishment opens new probabilistic outcome branches. Though they are not responsible for their crimes, we are not responsible for their punishment, and instead act in a manner benefiting society as a logical consequence of originally establishing and modifying that society in the first place, and our continued perpetuation of that society from then to now.
It is kinda funny to say that religion is incompatible with determinism when so many religious debates centered on the very concepts of predestination, which was itself an outgrowth of determinism as a consequence of an omniscient God. A God that can see the future and know our actions and their consequences thus knows, deterministically, our ultimate outcome, whether He defines it and thus defines free will, but that he determines it through observation and thus merely knows the consequences of our collective and individual wills in advance. More succinctly, omniscience was understood, accurately or otherwise, to imply determinism through knowledge at a bare minimum. As far as scientific theories of determinism are concerned, classical determinism, quite prominently in the form of Laplace's demon, does run solidly into the major issue of indeterminancy in quantum mechanics, which can grow into effectively-random macro-scale interactions. Probabilistic determinism is more interesting, but reliant on large-scale behaviors where quantum behavior can be averaged out, so to speak. It's also not nearly as certain, so to speak, as classical models; you can't say there is "one way" under such a principle, but simply a "most likely way."