It's a website that needs to be able to handle 3million visitors per day, with the majority of them being signups, or at least hitting the calculator. That's a lot of deep hits that can't be cached.
Then, add on a back-end that has to talk to insurance companies. These guys still have a tonne of Cobol code running around. There's nothing wrong with that (Seinfeld!), but I think it might indicate that their systems aren't necessarily built for online, real-time querying.
To recap, it is a multi-tier system:
1) Front end, performing user signup, and calculator.
2) Back end database. HIPA compliant, Sarbanes-Oxley compliant and able to deal with 100m customer records.
3) Feeds to remote systems, also HIPA compliant, Sarbanes-Oxley and other stuff.
So, you've got something that looks a lot like twitter (the back-end links), only more expensive because it needs to be Capital S secure, along with something that looks like an insurance company (the middle tier) and finally something that looks like a dot-com (front end calculator).
That's already a lot of hardware and software. "Free" open source doesn't actually save a lot of money here, since most of the money is in support (over 1/2 the 5year cost!). Now, triple it do deal with hot site failover, backups and other various disaster recovery plans.
Although they've had 3 years to get the system complete, the software was probably only developed in the last 10-12 months (at most). The rest of the time would have been spent in getting agreement on the data exchange formats with the insurance companies, deciding on a vendor to use for each part, and standing up an internal team to manage it. Then add in several parties involved playing schedule chicken with Congress, hoping for the whole thing to either be delayed or scrapped. Fun!
Finally, they went for a nationwide rollout for political reasons, which was guaranteed to result in peak traffic on day 0.