MPEG-1 quantizes both inter-frames and intra-frames, however I am curious why the quantization table for inter-frames are all constants, whereas the default quantization table for intra-frames is not.
This is because inter-frames contains only error residuals after subtracting predicted image from source image. Such residuals have almost uniform frequency distribution (if prediction ie. motion estimation is made correctly). For these frames we use uniform quantization matrices.
In contrast, intra-frames have very high power of low frequencies, they are hard to compress because of they enormous amplitudes. For these unpredicted frames we use low-frequency centered quantization matrices.