Thank you, Rick. This is getting more interesting. 3dcopy give the same error I got before:
3dcopy ${subj}.REML_IMs_reord+tlrc copy_reord.nii.gz
++ 3dcopy: AFNI version=AFNI_21.3.17 (Dec 30 2021) [64-bit]
** ERROR: nifti_image_write_hdr_img: NBL does not match nim

I forgot to say that in these cases a NIFTI file is actually produced. However, when I run 3dinfo on it, this is what happens:
3dinfo copy_reord.nii.gz
++ 3dinfo: AFNI version=AFNI_21.3.17 (Dec 30 2021) [64-bit]
** ERROR (nifti_image_read): short header read '/home/wgraves/data/bsafAG/data/mcwRA/copy_reord.nii.gz'
** FATAL ERROR: Can't open dataset copy_reord.nii.gz
** Program compile date = Dec 30 2021

But the 3dbucket command actually does work to successfully save it as a NIFTI file. Here's the output of 3dinfo on that resulting file:
3dinfo bucket_reord.nii.gz
++ 3dinfo: AFNI version=AFNI_21.3.17 (Dec 30 2021) [64-bit]

Dataset File: /home/wgraves/data/bsafAG/data/mcwRA/bucket_reord.nii.gz
Identifier Code: XYZ_IsUKqzK2tsZ7jPpyjmPHvw Creation Date: Wed Jan 5 17:31:16 2022
Template Space: TT_N27
Dataset Type: Anat Bucket (-abuc)
Byte Order: LSB_FIRST {assumed} [this CPU native = LSB_FIRST]
Storage Mode: NIFTI
Storage Space: 1,060,200,000 (1.1 billion) bytes
Geometry String: "MATRIX(2,0,0,-79,0,2,0,-79,0,0,2,-64):80,95,75"
Data Axes Tilt: Plumb
Data Axes Orientation:
first (x) = Right-to-Left
second (y) = Anterior-to-Posterior
third (z) = Inferior-to-Superior [-orient RAI]
R-to-L extent: -79.000 [R] -to- 79.000 [L] -step- 2.000 mm [ 80 voxels]
A-to-P extent: -79.000 [A] -to- 109.000 [P] -step- 2.000 mm [ 95 voxels]
I-to-S extent: -64.000 -to- 84.000 [S] -step- 2.000 mm [ 75 voxels]
Number of values stored at each pixel = 465
-- At sub-brick #0 'allwords#36_Tstat' datum type is float: -8.46264 to 8.56861
statcode = fitt; statpar = 673
-- At sub-brick #1 'allwords#400_Tstat' datum type is float: -4.15827 to 4.30151
statcode = fitt; statpar = 673
-- At sub-brick #2 'allwords#408_Tstat' datum type is float: -4.11308 to 5.13197
statcode = fitt; statpar = 673
** For info on all 465 sub-bricks, use '3dinfo -verb' **

----- HISTORY -----
[wgraves@argus: Wed Oct 13 00:43:56 2021] Matrix source: ; 3dDeconvolve -input 3Dout.words1+orig 3Dout.words2+orig 3Dout.words3+orig 3Dout.words4+orig 3Dout.words5+orig -mask epimask+orig -x1D R.xmat.1D -x1D_stop -polort A -censor Mcensor.1D -CENSORTR '*:0-5' -num_stimts 8 -stim_file 1 'MP.all.1D[1]' -stim_label 1 MP1 -stim_base 1 -stim_file 2 'MP.all.1D[2]' -stim_label 2 MP2 -stim_base 2 -stim_file 3 'MP.all.1D[3]' -stim_label 3 MP3 -stim_base 3 -stim_file 4 'MP.all.1D[4]' -stim_label 4 MP4 -stim_base 4 -stim_file 5 'MP.all.1D[5]' -stim_label 5 MP5 -stim_base 5 -stim_file 6 'MP.all.1D[6]' -stim_label 6 MP6 -stim_base 6 -stim_file 7 ./noise.1D -stim_label 7 noise -stim_base 7 -local_times -stim_times_IM 8 ./regressors/lsstimes.01.1D 'BLOCK(1,1)' -stim_label 8 allwords
[wgraves@argus: Wed Oct 13 00:43:56 2021] {AFNI_21.1.20:linux_ubuntu_16_64} 3dREMLfit -matrix R.xmat.1D -input '3Dout.words1+orig 3Dout.words2+orig 3Dout.words3+orig 3Dout.words4+orig 3Dout.words5+orig' -mask epimask+orig -tout -nobout -Rbuck 2426.REML_IMs -verb
[wgraves@argus: Wed Jan 5 16:35:24 2022] {AFNI_21.3.17:linux_ubuntu_16_64} 3dbucket -fbuc -prefix 2426.REML_IMs_reord.nii.gz '2426.REML_IMs+orig[1dcat reorder.1D]'
[wgraves@hermes: Wed Jan 5 16:38:29 2022] {AFNI_21.3.17:linux_ubuntu_16_64} 3dNwarpApply -nwarp MPRAGE_sag_al_qw_WARP+tlrc -source 2426.REML_IMs_reord.nii.gz -master /usr/lib/afni/bin/TT_N27_SSW.nii.gz -dxyz 2 -prefix /home/wgraves/data/bsafAG/data/mcwRA/2426.REML_IMs_reord+tlrc
[wgraves@hermes: Wed Jan 5 17:31:16 2022] {AFNI_21.3.17:linux_ubuntu_16_64} 3dbucket -prefix bucket_reord.nii.gz 2426.REML_IMs_reord+tlrc

So I guess I'll just stick with your 3dbucket trick (thanks for that!). But it would be nice to know what's going on here.

Best,
Will