The bugs, if serious enough, can make it not scientific, in the same way that a paper that has grave errors in it is wrong and thus obviously not scientfic. So, before we do a thorough review of the code, we should treat it with due scepticism.
EDIT: this can also be expressed in terms of risk analysis. For typical software, the consequence of a bug is low - most software is commercial, so the impact of the bug will be limited to that company's bottom line (with exceptions of software that can kill, but these people already are serious about bugs). Also, most bugs in most software are either highly evident (the button does not work, you get random segfaults etc.) or have limited impact.
On the other hand, the bugs in climate change models, given their "pipeline" nature (wrong result from one module is propagated downstream all the way to the final prediction of expected temperature change), can quite often have severe impact. They can also be not evident at all - they can for example change the final outcome predicition by 1 C.
Compound that with the fact these predictions are used to make trillion dollar decisions on global policies, and you can see that the actual damage done by bugs is not unlikely to be in the trillions. And that's why I say it's probably wise to subject the models to extreme scrutiny.