Oh, and one additional thing, related to section 3, number 5:

As I alluded to in my earlier mail, the first non-duplicate ack following a 
fast retransmit will update snd_recover.  This includes a partial ack, 
thereby breaking detection of further partial acks.  We should not update 
snd_recover on a partial ack.  (Which could also be stated as: snd_recover 
should never be retracted.)