Small deviations in One- and Two-Electron Energy

I discorved some issue with the output of SCF calculations. I was trying to reproduce the total energy from a molden file written by Psi4. Now the total energy was matching exactly, but there was some small error in the One-Electron and Two-Electron Energy, which cancels out when adding them together. Consider the following example:

molecule NH3 {
    0 1
    N  0.0000000000   0.0000000000  -0.0756322480
    H  0.9405596971   0.0000000000   0.3503377210
    H -0.4702798486  -0.8145485915   0.3503377210
    H -0.4702798486   0.8145485915   0.3503377210
}

set basis STO-3G
set scf_type pk
E, wfn = energy('scf', return_wfn='on')
psi4.molden(wfn, 'NH3.molden')

In the Psi4 output I get:

Nuclear Repulsion Energy =             11.7371760261232687
One-Electron Energy =                 -98.6231268695920988
Two-Electron Energy =                  31.4305310646644251
Total Energy =                        -55.4554197788044121

When recalculating these energies based on the molden file, I get:

Vnn  =      11.7371760261
E1   =     -98.6231278812
E2   =      31.4305320763
E_HF =     -55.4554197788

As you can see, the total energy matches up to the last digit, while E1 and E2 differ by 10^-6 Hartree. I repeated this calculation with Molpro and also tested the molden file created by Molpro: in both cases I get the second set of energies! So I my code to calculate the energies from the molden file should work correctly. I additionally compared the MO coefficient written to the molden file with those returned by the wfn object in Psi4, and they match.
Also this appears to be independent of spherical/cartesian GTOs or used symmetry.

Now the derivation is very small in this case, and does not influence the final energy. But this derivation increases with larger systems, e.g. switching to a cc-pvtz basis set in the above example results in a derivation of 3*10^-5 Hartree, which is already higher than the default SCF convergence threshold. I also have some other, more complex system which shows a deviation of 7 Hartree, but I am not sure yet whether this is really related.

So the question now is: Is this a known issue and related to some Psi4 internals or could this be a bug? In the later case it could maybe influence the results for larger systems.