Hi Martin,
I got the feedback from Kinetis product team with below feedback:
In this case, a 32-bit access is launched on a 16-bit port, that means there would definitely be two continuous accesses here.
In fact, S3 is used to terminate the transfer, so in this case, it is reasonable that the 1st access skipped S3 and goes to S0 directly, so it only take 3 clocks here.
But in the second access(it is writing here), S3 is a must to terminate the transfer.
We have a Flexbus timing about read from 8-bit port size device and write to 16-bit port device with 32-bit access (WS/ASET/RDAH/WRAH re all set to 0).
Please check attached picture for detailed info.
I zoom in the 32-bit write to 16-bit port device and get there take 7 Flexbus clock cycles, the first 16-bit write take 3 Flexbus clock cycles:

Wish it helps.
Have a great day,
Mike
-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------