Conclusions
For most spreadsheet applications where a standard normal deviate would be desired, unity variance with zero mean and a bell shaped distribution is likely more than sufficient.
For the generic use of the AMUD’s given here, 3 MUD’s per AMUD are probably more than adequate.
If multiple large sets of AMUD’s will be used, and it is only important that the sets will overall be statistically indistinguishable from a standard normal distribution, then at least 6 MUD’s should be used.
One advantage this method has over standard error propagation is the ability to see bimodal, skewed or other non-normal tendencies a system may possess even if the parameters defining the system are normally distributed (such would be the case with log-normal distributions).