One option is to use Ocean to run the nested MC and PA.
The GUI interface is craptastic for anything besides what
Cadence bothered to put in the can.
Another option is to simply create N Monte Carlo runs,
each at a different common mode voltage of interest.
Export the data from each to a file, pull them all into a
spreadsheet and grind the data as you see fit. This can
also give you 1/N the total solution time if you farm it
out to N different machines.
stddev may give you an answer even when it is given
non-gaussian-statistical data. What the use of that is,
might be nil. Certainly if you took the standard deviation
of deterministically generated points, it's likely to be
garbage in any statistical sense.