The "market for lemons" paper was not about the average quality of the market---it was about the selection effects generated when market participants cannot judge quality. Within a online labor market, there are platform-imposed features that reduce information asymmetries & make relative comparisons possible e.g., tests, feedback, verified records of billed hours. The informational problem is much worse without an intermediary, which is why these markets exist.
I agree with everything you say, but still believe that these sites are markets for lemons. The information asymmetries are still severe enough to compromise the overall utility of the market. (If they weren't, this thread would not exist.)
