I agree that they hate production quality code and numerical methods, but in some of these DFT codes for example (HF is only somewhat better), you fix some numerical bug and a bunch of theoretical results agreeing with experiment go straight out the window and with it your confidence in the computational model. These 'old farts' have probably seen this happen way too many times and on top of it from people who lack the basic theory knowledge, let along experimental, to point out a crap result from the code.