We'd be better off with a theory of everything designed by a neural network, at least people wouldn't belabor under the mistaken assumption that "theory of everything" actually corresponds to anything real.
Anyway, a string theorist on the faculty doesn't cost any more than a mathematician. For the others you are continually having to pay for LHCs, nanokelvin freezers, and optical tables.
That's exactly the problem - instead of trying to produce new theories, people keep researching string theory, in massive numbers (relatively to the size of the field of theoretical foundations of physics, of course).
Also, there are a few other interesting avenues to explore - coming up with a relativistic version of de Broglie-Bohm pilot wave mechanics (or as close as possible to get, given Bell's inequalities), for example.
If the string people were putting all their effort into coming up with a testable prediction, we might be better off. But they are mostly trying to show their untestable thing is better than the next one over, instead.