If you change the vector table settings to be "apparent states", the error goes to the millimeter range. Not sure why it isn't exact, but still a lot more accurate than the error margin of the ephemieris.
Vector table:
2440691.522916667 = A.D. 1970-Apr-15 00:33:00.0000 UTC
X =-2.592670760523030E+05 Y = 3.098416379718271E+05 Z = 1.914428047852615E+04
VX=-7.436303188457583E-01 VY=-6.130967807986689E-01 VZ=-7.420234845227733E-02
LT= 1.349132800160780E+00 RG= 4.044598383286231E+05 RR= 3.499474924335833E-03
Observer table:
$$SOE
1970-Apr-15 00:33 08 52 33.51 +20 22 27.6 -10.525 4.952 4.0445983833E+05 0.0034739 106.523 277.760 n.a. n.a. 29.577389 110.09186 0.2074268 n.a. n.a.
Difference:
1.3769022189080715e-06
Also, as I pointed out in my comment, the Center-Site name changes to GEOCENTRIC when you select the apparent states option, so it is likely using a different origin for the different types.