Thanks Andrew, these are useful references.
When Christian Buil wrote ISIS he recognised that once you subtracted the bias or offset signal, the remaining dark current in CCD cameras scaled linearly with exposure time. A dark frame taken with the longest exposure time could therefore be scaled to any other exposure time required. Hot pixels behave differently and are dealt with separately in ISIS. Exposures in spectroscopy can vary from 10 sec on a bright reference star to 600 sec or more on a faint target, so a set of darks at every exposure time with sufficient images per set to reduce statistical noise could potentially take hours to record. The approach taken in ISIS was a pragmatic solution. With temperature regulated cameras such a dark frame could have a relatively long useful lifetime.
With CMOS cameras having low readout noise per exposure and nonlinear dark frames, this approach may need to change.