Unfortunately we don't know what is the bug yet...
The main hypothesis was a hardware problem (communication between the FT231X and the TL16C550) but measure seem to indicate it was not the case.
Since that I don't get time to redo more software test
And I don't have hope for this week-end and this week...
For the CH376 part it's seem to work just fine, the DOS from OffseT work for me without problem. It's for the moment only in RAM mode ( I am not the only one without available time :/ ) but I currently use exclusively the SD for working after the boot.
For the serial interface, if we don't find the problem or can't correct it, I have a last chance workaround by using the serial interface only in 7 bits mode or having some escape sequence and forbidden value but it will have a very big impact on the transfer speed.