If you have enough I/Os left, use 8-bit bi-directional scheme. It is much faster than any serial communication. Here is what you may want to try:

1)8-bit bi-direction port (connecting uC A and B)
2)Req_A2B (request to transmit A --> B)
3)Ack_B2A (request granted for A --> B)
4)Req_B2A (request to transmit B --> A)
5)Ack_A2B (request granted for B --> A)
6)Strobe (connecting between A's EX0 and B's EX0)

Each uC will look at the other's Req line before asserting its own and then wait for Ack to be available.