Is there any data on packet duplication?

[Hal Murray <hgm+nanog () ip-64-139-1-69 sjc megapath net>]

How often do packets magically get duplicated within the network so that the 
target receives 2 copies?  That seems like something somebody at NANOG might 
have studied and given a talk on.

Any suggestions for other places to look?

Context is NTP.  If a client gets an answer, should it keep the socket around 
for a short time so that any late responses or duplicates from the network 
don't turn into ICMP port unreachable back at the server.  Nothing critical, 
just general clutter reduction.

I have packet captures from a NTP server.  I'm trying to sort things out.  
There are a surprising (to me) number of duplicates that arrive back-to-back, 
sometimes the timestamp is the same microsecond.  They could come from buggy 
clients, but that seems like an unlikely sort of bug.

-- 
These are my opinions.  I hate spam.