In a classification problem, I have c classes of data, each with e examples of this data, and each example is represented by a feature vector of length
Generally it's best to place the longest vector in a column. Therefore, (f, e, c) should be better than (c, e, f), assuming that f is indeed the longest dimension.
MATLAB is most efficient when working in columns, and many built-in functions are coded to work columnwise by default.
Source: http://www.mathworks.com/help/images/using-columnwise-processing-to-speed-up-sliding-neighborhood-or-distinct-block-operations.html.