> It kind'a bugs me a great deal that the networking code in NetBSD *can*
> allow errors to creep into a TCP circuit!

There being only 65,536 different TCP checksum values, there is a 1 in
65,536 chance that purely random data will successfully TCP checksum.

Of course, corrupted IP packets aren't (quite) random data.  The TCP/IP
checksum is fairly strong against certain kinds of errors -- but it is
notably weak against other kinds of errors.  As my little anecdote
showed, if you have a *bad enough* link, TCP's stubbornness inevitably
means that it passes some bad data along with the good.