I. From $\Gamma \vdash \neg\neg (a = b)$ and $\Gamma \vdash P(a)$, can we conclude $\Gamma \vdash P(b)$?
No. In the general case, $\Gamma \vdash P(b)$ does not follow from $\Gamma \vdash \neg\neg(a = b)$ and $\Gamma \vdash P(a)$.
Assume that we could conclude $\Gamma \vdash P(b)$ from $\Gamma \vdash \neg\neg(a = b)$ and $\Gamma \vdash P(a)$ for any context $\Gamma$ and formula $P$. Then in particular this would work for $P(x) \leftrightarrow a = x$. We'd have $\Gamma \vdash P(a)$ by the reflexivity of equality, so we would be able to conclude $\Gamma \vdash P(b)$, that is, $\Gamma \vdash a = b$. This would mean that equality is always stable under double-negation, i.e. that $\Gamma \vdash \neg\neg (a = b) \rightarrow a = b$ is provable in every context.
However, we can construct theories (models, topoi) where equality is not $\neg\neg$-stable, e.g. any model of Smooth Infinitesimal Analysis.
II. From $\Gamma \vdash \neg\neg (a = b)$ and $\Gamma \vdash P(a)$, can we conclude $\Gamma \vdash \neg\neg P(b)$?
We can always do this. The trick is to realize that double negation distributes over implication, even in intuitionistic logic.
Since $\Gamma \vdash a = b \rightarrow (P(a) \rightarrow P(b))$ is provable, we can introduce a double negation to get $\Gamma \vdash \neg\neg (a = b \rightarrow P(a) \rightarrow P(b))$, and distribute the double negations to get $\Gamma \vdash \neg\neg (a = b) \rightarrow \neg\neg P(a) \rightarrow \neg\neg P(b)$. Since we already know $\Gamma \vdash \neg\neg (a = b)$, we can conclude that $\Gamma \vdash \neg\neg P(a) \rightarrow \neg\neg P(b)$. Similarly, we already know $\Gamma \vdash P(a)$, from which it follows that $\Gamma \vdash \neg\neg P(a)$. Putting everything together we get $\Gamma \vdash \neg\neg P(b)$, which was to be shown.