Even without a crash, one of the transactions will be completed before the other ones.

If the coordinator crashes, one (or all) of the transactions will be stuck until the coordinator comes back - or, if it never does - until someone manually rolls them all back of commits them all.