There are simple nondetermintic procedures that can be implemented by digital circuits using arbiters that cannot be implemented by a nondeterminustic Turing Machine.
That is a strong assertion that requires proof. The consensus view is that there aren't. For example, one could claim that a TM couldn't simulate a coin flip as it cannot simulate true randomness, but this assumes that the coin flip is "truly" random without establishing it (which would be hard because of pseudorandomness). Or, in the case of arbiters, you could claim that the arbiter behaves like the magical collaborator in the stop/go examples, converting two analog inputs to a binary decision that takes an arbitrarily long time, but this only introduces yet another magical collaborator capable of producing analog inputs that are equal to arbitrary precision.
This is a common problem when we appeal to continuous natural phenomena, as their common description is usually a convenient, but imprecise, abstraction. Goldreich addressed this in On the philosophical basis of computational theories [1]: "A computational model cannot be justified by merely asserting that it is consistent with some theory of natural phenomena ... The source of trouble is the implicit postulate that asserts that whatever is not forbidden explicitly by the relevant electrical theories, can actually be implemented"[2]
Goldreich demonstrates the problem by showing that our abstractions of electrical circuits don't in themselves preclude circuits that violate computational complexity results or even decide halting, but that alone is insufficient to show that such circuits can actually be built.
There are certain formalisms that make this kind of error harder to spot: actor formalisms make it easy to hide impossible "computation" in a collaborator; some typed formalisms could hide computation in the syntax (which requires a lot or even infinite computation to decide).
This is a common problem when we appeal to continuous natural phenomena, as their common description is usually a convenient, but imprecise, abstraction. Goldreich addressed this in On the philosophical basis of computational theories [1]: "A computational model cannot be justified by merely asserting that it is consistent with some theory of natural phenomena ... The source of trouble is the implicit postulate that asserts that whatever is not forbidden explicitly by the relevant electrical theories, can actually be implemented"[2]
[1]: http://www.wisdom.weizmann.ac.il/~oded/VO/qc-fable.pdf
[2]: He adds "at no cost" because his focus is complexity, not computability