I have a similar story. The NSA hires entry level mathematicians for a pittance. They money they offered wasn't enough to raise a family, so I went with a job at a defense contractor that paid 50% more.

This "crating" thing you mention... I've never heard the term, but I've seen the effects. Some people settle down for a nice, steady, mind-numbing job in a SCIF because they know it's basically guaranteed employment for as long as the program lasts. You can spend your entire career doing the same thing for the same project, all while tidily walled off from the world because of security constraints. My peers and I joke that it's welfare for engineers.

This article is idiotic.

How much latency did the network have? If latency was anything more than minimal, window scaling (disabled by default in XP) would allow the Ubuntu machine to better fill the connection.

How much packet loss, if any? Selective ACKs and Timestamps can affect recovery time after lost packets and transmission timeouts. XP does not use either option by default.

A quick look at /proc/sys/net/ipv4/tcp_congestion_control indicates that the default on the ubuntu machine where I'm typing this is the "cubic" flavor of TCP congestion control. I believe that XP uses the "New Reno" variety.

All of these can have an impact, and obviously TCP can be tuned on either OS. The discrepancy is probably due to TCP options that are relevant to his connection (and configurable in XP), but disabled by default.