Certainly, it is even longer; if you move the date of comparison 6h or so forward the discrepancy will increase. Also, the differences between various types of astronomical year are much smaller than the current effect.

I got a formula for earth's mean anomaly:

M = 358.47583° + 35999.04975° TJ - 0.15° 10^(-3)TJ^2 - 0.33° 10^(-5) T_J^3

which says that M should increase by 35999 degrees in a Julian century, i.e. in 36525 days - except for small non linear perturbations (from J. Meyer on sundials, who has it from Dumoulin' book on practical astronomy, who in turn has it from Newcomb). That's what I expected.

It seems that Mathematica's data are just routed through from JPL at Caltech, and I posted my question to them. They replied rather elaborately, and I have not yet completely digested their answer, but the explanation may be that formulas as the one above do not really describe earth, but the common center of mass of the earth-moon-system. Such effects could indeed be large enough.