It seems to me, that this "rule" is one of sources of the '51-planet popularity, probably due to the enhanced feeling of compatibility, i.e. that if needed, one can easily switch to a different derivative.

if we discuss the rules, will they still be unwritten? :)

"Direct compatibility" is not really the issue, althought that would be nice. The issue is more that if you have to peruse the datasheet for all kinds of things that you otherwise "know" the developement is extremely error prone. Had I not detected the "irregularity" of the F3xx series (by someone else posting that he got burnt!!!), I would have designed a board with clock input to XTAL1. Of course when I then realized that I also had to recode the whole IIC processing, and a whole lot more, I chose another chip.

Adding a SFR for some function is, in my opinion, not breaking a "rule", thus many derivatives with "advanced features" are not rulebreakers. Just look at the SILabs F12x series with the myriad of SFRs, but still capable of running a standard '51 code with minimum adaptation effort.

so

Are there any other derivatives "violating" this "rule"?
I know none

Erik