I am more than honored to receive an email from the boss personally.
The discrepancy between the iom328p.h and the ATMEL datasheet obviously has existed since the very beginning, and it seems that it will never be abandoned for reasons of compatibility.
Before I sent my posting, I had been looking everywhere for the discrepancy, but couldn't find any clues.
I came across that discrepancy using the avr-objdump.exe in order to understand what code gcc is producing and sometimes to optimize it using the inline assembler.
For all users who, like me, are using the avr-objdump.exe, a reference to this discrepancy would be very useful to compensate for their confusion.
In the meantime I fixed a table next to my monitor telling me both vector numbers ...