OK, I did a first measurement regarding noise and symmetry. As expected, the noise on the output of the DC/DC (and thus on +Vcc/-Vcc) is quite substantial. The attached screenshot was taken with the output set to ~35V and the input at 5V (linear bench supply). There are two independent disturbances. One is triangular in shape and has a period of a bit more than 5µs (~190kHz) with a peak of around 150mV. The other one is a (somewhat) symmetrical glitch with a period of rougly 10µs (100kHz). In sum, the RMS noise is about 27mV.
[EDIT]
The 2nd (symmetrical) glitch has its origin somewhere else, not in the module. I currently blame my UPS but have to investigate further.
Still, the noise on the DC/DC module is far worse than on others I have with a (fake?) LM2577. Guess the problem is that this module doesn't have any LC filter at its output.
I then measured the voltage between the outputs to check for symmetry. With the DC/DC set to 34V (i.e. 34V measured between VCC+ and VCC-), the virtual ground was slightly off. VCC+ to VGND measured 17.03V while VGND to -VCC measured -16.97V. So there is a +/-30mV deviation at 34V. Still, 0.03V/17V*100% equals 0.176%. So the symmetry is pretty good.
Note that all measurements were done without load other than the onboard LED and the TD2030A. With 34V on the DC/DC output and a 10k resistor, the current through the LED should be 3.4mA. Additionally, the quiescent drain current of the TD2030A is supposed to be around 40mA. On the input, there were 348mA going into the circuit at the time. So the efficiency of the DC/DC for very low output currents should be something like
(43.3mA*34V)/(348mA*5V)*100% = 84.8%.
There's a bit of guesswork involved since I can't really measure the quiescent current drain of the TD2030A but I guess an efficiency of 80-90% seems realistic.
Besides, I guess we can assume this as a worst case scenario (low input voltage, high output voltage, low output current). So I did another test with the input voltage at 12V. The DC/DC output was 34.08V at this point and the current flowing into the circuit was 144.4mA.
(43.38mA*34.08V)/(144.4mA*12V)*100% = 85.32%.
Again, note that this is the efficiency of the DC/DC board only, since I assumed the quiescent current draw of the TD2030A and the LED as loads. Obviously the efficiency of the whole rail splitter circuit including the DC/DC will be catastrophic for small (external) load currents. Only if they are significantly higher than the internal load current (LED + TD2030 quiescent current), it will be possible to get into the ~80% area again.
Side note: when I first played around with the board today, I observed significant flickering on the LED. I first used USB connectors, then moved to the bench supply. But the flicker didn't go away.Only after some time, it disappeared and didn't appear (yet) again. I didn't observe and voltage fluctuation on the outputs though (other than the noise described above).