Actually, gaydar has been held up with statistically significant results, but the effect isn't very strong; people are only slightly more likely than chance to be correct in their hunches (so if the experiment is structured so that the pictures people are looking at are 50-50 straight-gay they might get perhaps 57% right). The effect does not seem to be contingent on how long someone looks at the pictures, so people can tell at a glance.
I'm pretty good at it.
Bisexuals are imposable to detect because most of them are "str8 acting" in public settings.