I was told it would be fixed in ESX 5.0 and it is. I have a vm (centos 6.0 x64) that routes IPv6 traffic on esx 5.0 and it works without workaround (i.e. lro off).
On the other hand I had trouble with upd traffic witch centos 5.7 x32 machines on ESX 5 and I had to do the following
to avoid getting "bad udp chkum" (with tcpdump -n -vvv)
ethtool -K ethX tx off
otherwise the tftp server would not work...