It's worth noting that CGI was also responsible for the ill-fated long gun registry in Canada, which was recently canceled due (in part) to massive cost overruns [1] and inability to cope with large numbers of users.
As an US RKBA activist since the early '70s (sic), I watched it from a distance in horrified fascination, and the Wikipedia section you cite matches my memories, that while the IT component was nasty and much bigger than expected/hoped/etc., it wasn't what crushed the registry. The job, including most especially human processing effort, was just massively larger than anticipated and was therefore not even vaguely covered by the statutory fees.
Yeah, this was a good lesson how not to manage a large IT project. They didn't really understand the requirements or anticipate the user load, but the cost estimates were WAY off - from an anticipated $2m annual net cost, the actual net cost ballooned to $66.4m for 2010-2011. To be fair, the federal government was also at fault here but this absolutely blows my mind.
I guess it depends on the tradeoffs of reputational damage and making money. I gather the government told them "build a system that can handle 50,000 to 60,000 simultaneous users", which was "based partly on the all-time high of 30,000 simultaneous users for Medicare.gov" presumebly during the Plan D enrollment season (although per the article the theoretical max of healthcare.gov has not been disclosed: http://www.usatoday.com/story/news/nation/2013/10/05/health-...).
Whatever the company thinks of the official goal, there's strict limits to how much extra they can spend to handle a much larger goal, vs. their getting paid on an emergency basis to bulk up the site....