Here is an analysis of how various factors including engineering improvements could see a

Here is an analysis of how various factors including engineering improvements could see a system go from 6 months to 18 hours of execution time. Not a classic case of a failure based post-mortem because there was nothing at stake here but it does give you insight into the details of how many small improvements can give an overall large improvement.

Lots of deep technical info on that blog. See also the docs for the fractal program:
http://www.cygnus-software.com/docs/html/64bitcalcs.htm