I clearly understand the situation you are describing.
I'd hope that the first slave device would ignore the transaction with the second slave device (that's what I2C addresses are for, right) and thus keep its internal state machine (unaffected by the transaction with the second slave). But I don't know for sure whether I2C supports these kind of "piecewise" transactions.
Therefore I wonder why there's no reply from NXP. I'd be happy to have an official view on this situation.