if I put zeros at the end of an input sequence, and take fft, then take ifft of a greater length by appending zeros to the fft output, will the zeros of the input sequence reflect on the final ifft output?

I tried to do this in MATLAB and python, it seems the last part of the final IFFT output deviates from being nearly zero in the absence of any zero-head in original data.

However, when repeating the same process for DCT and IDCT, I found that the last part of the output is smoothened to nearly zero.

I am taking 512 point FFT/DCT and 2048 point IFFT/IDCT for that

Could anyone give any mathematical explanation to this?