The quantisation matrix to encode the error frame contains 16 in all locations. Error frames may contain isolated details and hence high frequencies may be important
The MPEG standard allows implementations to change this matrix and this is often done