The V&V guidelines are certainly a starting point, but they are there only to indicate the range of experiments that CFAST has been compared to; not to provide an actual limit to the calculation. With additional data, the V&V range is likely to increase. There is no hard rule for an upper range since it's dependent on the purpose of your individual scenario. In a long corridor, it takes some time for an actual layer to form and there may be a profile of temperature down the corridor. For detection scenarios, this may be important (hence the inclusion of a corridor ceiling jet algorithm in the model), but for fully-developed fires, the two layer assumption holds pretty well.
For the corridor options, I'd suggest modeling with and without to see if there is a significant difference in your results. For the shaft option, it's intended for tall and narrow compartments like elevator shafts where mixing deer to wall interactions limits the formation of two distinct layers.
Rick Peacock