I suspected that the air leaking from the legs might cause noise in acoustic bandwidth, so I measured the beat signal with unfloated /floated table. The beat signals from the two look similar. Air leaking from legs does not cause any extra acoustic coupling from 100Hz - 3kHz in the beat signal.
fig1: Beat signal between floated and unfloated table. There is no significant difference between the two.
On Friday evening, I turned off the air compressor. When I came back on Sunday night, the table was not floated. I measured the beat signal, then turned on the compressor, realigned the beam, then measured the beat again. The results were similar in 100Hz- 6 kHz band. Floating the table with these leaking legs will not add any extra acoustic noise to our signal (at least, at this level).
Since the result looked nice with fewer mechanical peaks at night, I took a chance to check if it was limited by frequency noise of marconi in PLL loop or not. I changed the input range from 1kHz to 200Hz (their frequency noise level should be different and observable at 100-1kHz bandwidth, see here), but there was not different between the two input range, see fig2 below.
fig2: beat with different input range. There is no significant change in the results.
It is likely that the flat level we are sitting on are detection noise + shot noise. This will be carefully checked next.