once, again thank you rick,

It works now, but its very demanding with the inputs:

if anyone wants to try this in the future (with my files provided above), this is the syntax:

3dConvolve -input zeros_template+orig -polort -1 -num_stimts 1 -stim_maxlag 1 20 -stim_file 1 4blocks_zeros_ones_for_convolution.1D -iresp 1 mean_iso_HRF+orig -output Ideal_Block_Model

(the IRFs have 21 time points so as David figured out, stim_maxlag has to be set to 1 20)
Also very important, the 3d+time template consisting of zeros has to be created, with the same voxel size and time points as the final result.

Maybe there is a less tedious way, but what I did is, 1. cut out a single sub-brick with 3dTcat, 2. multiply it with 0 using 3dcalc, then 3.
loop over 3dTcat with the -glueto option, to create in my case the 710 repetitions.

set x=1
while ($x<710)
3dTcat -glueto zeros_template2+orig zeros_template+orig
@ x++
end

(as said very tedious, maybe i just missed a simple option in 3dTcat that can do that).