"It is only suitable for uni-dir synchronous HW automata."

No. This is not entirely true. Do you think your PC's printer port works in synchronous mode?


"It is because of its guaranteed single-cycle timing. I can't be sure that a slave MCu has accepted data dyring a cycle."

You concern applies to serial communications as well! You need to know how many cycles (T) the receiving end takes to store the data from the moment it sees the strobe in the worst case scenario. You can assign the highest priority level to this interrupt. On the transmit end, you will have to do this:
1.Put data on the bus
2.Toggle strobe line
3.Wait 2T
4.Step 1 - 3

"In addition to acknowledge line, I also need an interrupt line for master and slave to inform each other that they have some data. "

If you use Master/Slave (or Poll/Response) scheme, Master will always be the one who initiates the communication. The dialog goes like this,

Master: Do you have data for me?
Slave: No or Yes, here you go...
Master: Got it! or No, please re-transmit
(three-way handshaking)

Master: I got some data for you
Slave: Got it or No, re-transmit please

Both Master and Slave shall time out their waiting peroids if the expected response does not come in time.

Think how you would do with a serial communication, except this time you just have to do it parallelly.