RCKMPI is MPICH2 with SCC specific "channels". Channels in MPICH do nothing more than byte transfers. Everything else is the same as MPICH2, including the behavior you are asking about.
Bytes to transfer are stored in a FIFO and are transferred in order (the only exception being shared memory communication with Nemesis, but never with RCKMPI channels).
So to answer your questions:
(a). B will block until both messages are received in order to receive the one tagged T2. The one tagged T1 will be stored in an internal buffer until the program does the corresponding receive.
(b). The reception is guaranteed in any case, it is just a matter of latency given the in-order transfers.