Is there a faster way to average datasets than using 3dcalc?
Yes; program 3dmerge with the -gmean option should execute faster and can also deal with more than 26 inputs. Also, program 3dMean does essentially the same thing, and is specialized for this purpose (so has many fewer options to wade through than 3dmerge). 3dcalc uses a general purpose expression parser that I wrote for other purposes (part of an imaging radar simulation) in Fortran, back about 1984. It isn’t very efficient compared to compiled C code, but it is handy for “one off” calculations.
