Principle A DOES account for such sentences: in '*Himself saw John' the anaphor is free (= unbound) in its binding domain and the principle is violated (as well as Case Theory), but, in your example 'It is himself that John is seeing', 'himself' is coindexed with the phonetically null 'wh'-operator hidden in the Spec Comp of the relative 'that'-clause (recall the old 'doubly filled Comp Filter') and also with the 'copy/trace' that the 'wh'-operator itself leaves after the verb 'seeing', which IS correctly c-commanded and bound by the subject 'John'.
The apparent trouble arises only if it is forgotten that the 'links' of any given 'chain' (e.g., antecedents and 'wh'-items and their traces, in this case) must contain the same referential specifications, which, for the purposes of Binding Theory, makes them function as a unique 'discontinuous' category. Hence, if the trace/copy of a 'wh'-item is bound in its referential interpretation by a suitable antecedent (= 'John', here), so is its null 'wh'- 'copy' (= antecedent) in Spec Comp itself, and since relative 'wh'-items must themselves have a co-referential antecedent (= himself' in this case), by definition, the antecedent 'himself' must refer to the same individual that the 'wh'-item and its trace/copy refer to, ergo it IS bound by the antecedent 'John' as Principle A requires.
Similar Binding Theory pseudo-problems arise in cases of topicalization like 'About himself, John almost never says anything __': Having been displaced leftwards, at 'surface structure' 'About himself' seems to be neither c-commanded nor bound by 'John', but its trace/copy is, so such sentences do not constitute violations of Principle A, either.
