I'm disappointed their judging environment is a single core. Multiple cores are becoming increasingly important these days, and with more and more functional languages supporting parallel programming, it's a shame that those languages won't be able to show off their full capabilities on this highly parallelizable task.
Weird to see the cost excuse in there, since the cost of judging 100 teams on 100 maps for 3 minutes each on 1Ghz 1GB-RAm machines is 30,000 minutes = 500hrs < $50 on EC2 small machines