fft of large 3d volume

I am using the new partial read of a matfile to do a 3d fft of a large volume, as it is too large to fit into memory. I am testing on a smaller volume to start with. First I fft the volume (matObj.X) slice by slice, then I do an fft in the remaining dimension in a second pass and put the output into a temporary matfile. The first pass takes about 2 seconds and the second pass takes about twenty minutes. I assume the second pass takes so much longer because the data in the final dimension is not in a sequence on disc. Is there any way to speed up the second pass? Code is below, I hope it is understandable! Thanks in advance. Chris W.

tic
wbh = waitbar(0,'fft first pass');
for i = 1:sz(3)
    slice = double(matObj.X(:,:,i));
    fftOut1 = fft2(slice);
    localMatObj.x3real(:,:,i) = real(fftOut1);
    localMatObj.x3imag(:,:,i) = imag(fftOut1);
    waitbar(i/(sz(3)), wbh);
end
close(wbh);
toc

tic
wbh = waitbar(0,'fft second pass');
for i = 1:sz(1)
    for j = 1:sz(2)
        fftOut2 = fft(complex(localMatObj.x3real(i,j,:),localMatObj.x3imag(i,j,:)));
        localMatObj.x3real(i,j,:) = real(fftOut2);
        localMatObj.x3imag(i,j,:) = imag(fftOut2);
    end
    waitbar(i/(sz(1)), wbh);
end
close(wbh);
toc