I've built some generic functions in an "Aggregations" dimension to return
the Max, Min, Median, etc. for the current slicing for whatever measures are
displayed. The general syntax is as follows:

median(
nonemptycrossjoin(
descendants([DateTimes].currentmember,[DateTimes].[Hour],Self),
descendants([Iteration].currentmember,[Iteration].[Iteration],Self),
{[Aggregations].&[1]}
)
)

I want the functions to find, in this case, the median value across hours
and monte carlo iterations for the current slicing and measure(s). The
functions work, but are way too slow. The functions are in a virtual cube,
and there are virtual dimensions in the underlying physical cubes. There
are 8-20 million rows in the virtual cube set depending on the study.

Is there another approach that might yield the same functionality, but will
perform better?

Thanks
Jim