Thanks Michael, I also noticed these similarities and differences when
looking through your file.
I was also told that the I2C controller in the 89C66x and C554 is exactly
the same (except for the clock generation), and the C554 data sheet
contains additional detail information.

That DS clearly mentions that when sending a byte, that byte is
also clocked back into the shift register again (from the SDA line), so
afterb having sent a byte this byte will be found in I2DAT. This is exactly
what I noticed, and why all my data bytes are one byte late.

What I still miss is some information about if this receiption of a sent
byte also forces some SI/I2STAT action.