They're still initially scaled in the same way (i.e. based on the ability of the candidature, as defined by their performance across all courses), but the distributions are carefully scrutinised afterwards. It's pretty hard to scale a course with only two students without them ending up at opposite ends of the spectrum, and sometimes the non-linear transformation doesn't converge anyway, so in these cases a simpler linear transformation would be used to scale the marks.
However, that's only the process by which the marks are scaled - the degree of scaling itself (i.e. the scaled mean) is still determined using the same method as is used for other courses. It's the only objective way.