Earlier quoted context omitted.
I guess you never tried to find optimal join orders for you sql engine.
I actually have :), it was a time travelling database for $work. In our case, we assumed that queries wouldn't change too much (applications running on them tend to query the same tables over and over again). Thus we just picked an order "at random" (though IIRC, there were some heuristics) for table-sets, otherwise we used the best performing one seen so far -- but it would randomly run another query with a differen…
A random search like that might give okay results but has no guarantees approximation ratio.