Should I use equal area or equal angle projection?

The choice of projection (Equal Angle or Equal Area) has no effect on the calculated mean plane for sets.  The mean plane calculations are done using the raw data on the reference sphere.  The projection type affects the display of the 2-dimensional stereonet, but does not affect the calculation of the mean plane of a joint set.

If you switch between Equal Angle and Equal Area, using the same set windows, you should get identical results for the mean planes.  

Regarding Equal Angle versus Equal Area – as noted on page 3-18 of Dips Tutorial 3;

When making visual estimates of clusters and variabilities, it is actually more appropriate to use Equal Area projection to reduce areal distortion and improve visual estimates.

Since using Equal Area preserves areas (i.e. relative areas on reference sphere versus projection), this means that there is no distortion of area from the center of the stereonet to the perimeter, therefore allowing you to more accurately compare sets anywhere on the projection.  So you might consider Equal Area to be more suitable when you are initially determining the extent of sets and their relative importance.  Of course, the Fisher concentration also determine the ranking of sets.

If you are performing kinematic analysis, then Equal Angle projection is more appropriate.  Since angular relationships and shapes are preserved (e.g. small circles remain circular anywhere on the stereonet), you can more accurately assess kinematic stability with an Equal Angle projection.

In general, the relevance of Equal Angle versus Equal Area was probably a more important issue in the days of manual stereonets, when it took a great deal of time to generate even the most basic plots.  Furthermore, since detailed post-processing (e.g. wedge stability analysis with factor of safety) is now handled by other software packages (such as Swedge and Unwedge) the issue of stereonet projection is of even less importance.