Software is half art and half science... developing a truly new game is probably more art than science. If the architects go too far, they end up biting off more than they can chew - and there becomes a high risk of failure... the trick is to balance things so that the game is good/new enough that people will buy it, and that it can be developed in a reasonable amount of time so that revenues can be generated before the company's coffers run dry (Stardock offsets this phenomenon somewhat buy allowing folks to buy in at beta time, and by having more than one product).
The fact is, developers/architects CAN'T AFFORD to let things slip too much, otherwise they'll put themselves out of a job. Having said that, some slip is ALWAYS going to happen - if it didn't the game would probably be a POS in the end.
In any event, schedule is only 1/3 of the entire equation... resources ($), and quality being the other legs of the table. Many times, it is best to slow things down to save money and increase quality.
Another thing that affects schedule is whether or not the company is able to keep key people - if a key person leaves at a critical moment, the schedule can be totally screwed... if enough key people leave, the whole house of cards can collapse. So - the trick to being a successful company is to recognize and reward high performing individuals... and the trick for high-performing individuals is to work for companies that are good at doing this