GAMES has about 800 participants much larger than any theoretical computer science conference I've ever attended (which was STOC 1987 in New York at about 500). Are there that many more game theorists than CS theorists? No, just that GAMES is held only every four years and has massively (up to 14) parallel sessions where most everyone who wants to talk can talk. This gets the whole community together, much like how Mitzenmacher describes ISIT, an information theory conference. The invited plenary and "semi-plenary" (5 parallel sessions) talks at GAMES are reasonably strong invited talks, though the regular sessions are more mixed.
So should we have a TCS Congress held every n years with theory broadly defined and massively parallel sessions to encourage participation to bring our community together at least once in a while? (And no, FCRC doesn't count.) Unless we have a corresponding reduction in emphasis of the other conferences, having one more conference to attend will not likely have the desired effect.