You are quite correct, small pieces of 235U are stable. The difference is that low concentrations of 235U in natural uranium (because of it's faster decay than 238U) make it harder to get to critical mass, even with chemically pure (but not isotopically pure) uranium. IIRC, reactor grade is around 5% 235U, while natural uranium is 0.7%. IIRC, pure natural uranium metal, at least by itself, doesn't have enough 235U to sustain a chain reaction, even in a large mass. (but I vaguely recall that the original reactor experiment with just the right spacing of uranium metal lumps and graphite moderator may have been natural uranium - I need to check this... (short of time right now)) (I'm still not quite sure - Chicago Pile-1 is documented here but the web page described the fuel as "uranium pellets". I think they mean natural uranium, in which case I withdraw my statement that isotope separation is a prerequisite for nuclear power.)