In ADE XL (assuming you're using IC61X; there's no UI support for doing what you want in IC5141), on the "Gear" icon next to the Monte Carlo Sampling cyclic choice (also available via Run->Monte Carlo Sampling in the pulldown), there's a button at the bottom of the form called "Specify Instances/Devices". If you bring this up it allows you to specify devices for mismatch - either as an inclusion or exclusion list. It's partly used to allow the mismatch to correlate between multiple tests which have different testbenches, but can also be used for the purpose you describe.
The mmcflag parameter that Shawn describes must be specific to a particular PDK because it's not a parameter in spectre as far as I know (certainly not for bsim4 or bsim3v3; to be honest, the mismatch variation is handled outside of the model, so I can't see how it would be done in the device model. For a PDK they might choose to multiply the mismatch variations by this flag inside their subckt models, but as I said, that would be PDK-specific).