The concern that Steve Gibson has is that machines running XP are more likely to be compromised by Sub7-style trojans, and so the raw sockets ability is hence more dangerous.
I'd respond to this by making the point that whilst versions of Windows prior to Windows 2000 don't support the IP_HDRINCL socket option, they do still provide mechanisms for spoofing IP packets (you can inject them through NDIS/TDI, which are lower-level parts of the networking driver stack).
That is to say, even on Windows 9x -- an OS that has no security mechanisms whatsoever (Win2K/XP restrict SOCK_RAW to Administrators) -- it is possible, and has been possible for a number of years, to spoof IP headers.
So he's being paranoid, reactionary, and quite ridiculous in his casting of blame.