but.... you stillonlyhave two predictors: machine and person, each with multiple levels. i have trouble thinking that the level of a predictor and the predictor itself would be the same thing.

perhaps you're getting confused and mean that multilevel model (linear mixed models) can handle unbalanced designs more easily? and you don't have to worry about which sums of squares decomposition is more optimal given the design?