I haven't exhaustively looked at all cases, but it appears that, at least for the "bad data" case 05:10-06:10 AM on 5 June at Platteville, dew formation caused the bad h2o and presumably co2 values. During this case, RH is at least 97% and w'h2o' fluxes during the preceeding hour are negative (water deposition to the surface).
I note that this occurred despite the Li7500 being mounted at 45 degrees upside down, but presumably dew on the mirror also would affect the signal.
(This analysis was assisted by David Reed.)