ccx_SPOOLES.exe result better than ccx_PARDISO.exe result

Just starting with CalculiX, and I obtained a result which gave me pause. Perhaps it is related to the expansion of T2D2 elements, but I could not explain the observation

Using ccx_SPOOLES.exe and ccx_PARDISO.exe, different results are obtained from a little test script.
The script was an Abaqus inp file from the internet (https://www.seas.harvard.edu/courses/es128/Truss3.inp), modified to run with ccx.
CCX for windows was obtained from http://www.dhondt.de/calculix_2.17.1_4win.zip.
I installed the Intel libraries, and updated my path.
I monitored the displacement of node 4, and expected the Z displacement to be near zero.
This was close to true for SPOOLES, but not for PARDISO.

SPOOLES:
-1 1 0.00000E+00 3.38813E-21-2.54110E-21
-1 2 0.00000E+00 0.00000E+00 8.47033E-22
-1 3 0.00000E+00 0.00000E+00 4.23516E-22
-1 4-7.05766E-03-2.84157E-03 8.76424E-04

PARDISO:
-1 1 0.00000E+00 0.00000E+00 1.99476E-19
-1 2 0.00000E+00 0.00000E+00 1.65171E-20
-1 3 4.23516E-22 4.33681E-19-4.23516E-22
-1 4-7.05766E-03-2.84157E-03 -5.32285E-01

My windows path is
c:\intelpython3;c:\intelpython3\Scripts;C:\intelpython3\pkgs\openssl-1.1.1g-vc14he774522_5\Library\bin;C:\calculix_2.17.1_4win;C:\calculix_2.17.1_4win\cgxCadTools;C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2020.4.311\windows\ipp\lib\intel64_win;C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2020.4.311\windows\redist\intel64_win\mkl;C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2020.4.311\windows\mkl\lib\intel64_win;C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2020.4.311\windows\mkl\lib\intel64;

Stripped of most comments, the script becomes:

*HEADING
Three Bar Truss: ES128 Example Problem
*NODE, NSET=BARSET
4, 10.0, 0.0, 0.0
1, 5.0, 10.0, 0.0
2, 7.5, 10.0, 0.0
3, 10.0, 10.0, 0.0
**
*ELEMENT, TYPE=T2D2, ELSET=BARS
1, 4, 1
2, 4, 2
3, 4, 3
**
*SOLID SECTION, ELSET=BARS, MATERIAL=MAT1
0.1
*MATERIAL, NAME=MAT1
*ELASTIC
30e6, 0.3, 294
**
*BOUNDARY
1, 1, 3
2, 1, 3
3, 1, 3
**
*STEP, PERTURBATION
*STATIC
**
*CLOAD
4, 2, -1000
**
*OUTPUT
**
*EL PRINT, ELSET=BARS
S, E, COORD
*NODE PRINT, NSET=BARSET
U
*NODE FILE
U
**
*END STEP

At first glance it reminds me of this (topic from yahoo groups):

“*RADIATE wrong solution and only a single iteration x x”