I solved it by declaring the array in the object structure like this:
float x_iem_coeff[5][2];
and in the new function I create another, one dimensional array, and run a nested for loop and set the local array's elements to x_iem_coeff[5][2] one by one. This seemed to work, no errors during compilation. The object crashes Pd when the DSP is turned on, but I guess (and think) it's something else that causes that. The array declaration and initialization shouldn't have anything to do with the DSP state, right?