Here the encryption is done as follows : $$C=P^e \textrm{mod} \, n =(P_1||P_2)^e \textrm{mod} \, n.$$ Here's my scenario that the adversary with CCA2 wins.
The adversary chooses $X_1, X_2$ of the same size with $P_1, P_2$ and multiplies $C'=(X_1||X_2)^e \textrm{mod} \, n$ by $C$ to get $CC'=(P_1X_1||P_2X_2)^e \textrm{mod} \, n$. The adversary asks the decryption oracle to decrypt $CC'$ to get $(P_1X_1||P_2X_2)$. Then the adversary can get $(P_1||P_2)$ by just multiplying the inverse of $(X_1||X_2)$. Finally, the adversary can decrypt the original message as Bob does.
This is the same as CCA on textbook-RSA. I see that RSA-OAEP is IND-CCA2 secure, but why is this attack impossible?