What kind of concurrency issues were you running into with a RDBMS that has atomic transactions?

Are we talking about distributed store situations where CAP theorem limits apply, or something else?