First of all, workloads more often than not vary over time. A variation in computation necessary to serve user facing request is usually caused by diurnal cycle. At best such a difference between daily peak and trough is 2x. However there are business cases where variation is extreme.
Conventional Cost Modelis not flexible enough
Suppose you need a data management system in your computing infrastructure. Approaching this with conventional technology usually involves constructing a benchmark for specific software and understanding hardware that needs to be provisioned. Thus hardware acquisition, maintenance and software license and support costs create total cost of ownership of the solution. There are a couple of painful problems with this conventional approach.