The absolute threshold of detection (e.g. pmol, nmol, etc...) for a protein depends largely on the properties of the primary antibody and the method being used to image the blot. It depends on how well the primary antibody binds to the protein in question at a concentration that doesn't produce undue non-specific, secondary bands.
With regards to detecting a protein in a cell lysate, it will further depend on how abundant that protein is in the type of cell being analyzed. As a result, the absolute number of cells needed to reliably detect a protein depends on the expression profile of that protein.
Most Western Blots that I have seen published use somewhere between 5 and 50 ug total protein. If your antibody is any good at all, 20 ug should be more than enough protein for detection if it's expressed at an appreciable level. As others have said, 1 million cells and even 200k cells should contain more than enough protein to load this amount.