It's a little extra code to write, but it should be able to be made generic -- follow the lead of @Cris LaPierre in that thread; wrap the categorical argument in your plotting routine inside removecats and use a logical addressing vector to address the elements left after your selection.
For the above example data;
E=categorical(C(:,1));
M=[C{:,2}];
ix=(M<=8);
scatter(removecats(E(ix)),M(ix))
results in the same figure as yours above as far as data but without the empty categories (and using plain vanilla scatter, of course).
If you're programmatically setting the limits or other selection criteria, it should be feasible to compute the required indexing variable(s) programmatically and avoid the manual intervention.
