I have found the same problems in simulating capacitive DC-DC charge pumps - that is the input power is less than the output power. But changing the simulator from Spectre to BDAsim gave me a different answer - with the Pin > Pout.
Make sure that the pulse generation does not contribute to the output power - the phase1/phase2 'vpulse' should have high-impeance loads!
If you use ideal switches and change the on-resistance of the switch you'll find that the simulator will give other answers - sometimes wrong, somtimes appearing correct. Ideally you should get near 100% efficiency with ideal components.
I looked at the averaging of the input current and output voltage - no errors from the waveform post-processing. So I can anly assume that its due to the simulator solver not getting the solution accurate - trying different tolerance settings and/or timestep controls may help.
Good luck...