I'm using the Iso Cluster Unsupervised Classifcation to take a CIR aerial and find the impervious cover. I'll tell it to use 6 classes but on occasion my output will only have 4 or 5 classes.
This is a step within a larger model so my solution needs to be general and apply for any input raster.
This is a feature of the isocluster algorithm -- the number of classes you get is dependent on your input data. If the process can't find six distinct enough classes in your input data, you will get fewer classes than the number you ask for.
You may be able to get more consistent results by modifying the Minimum_class_size pararameter. This will allow a class with fewer cells.