Ah, that's the classic debate really, though none of the (many, many) studies has really found much difference between the two options. The argument goes that while potentially hot, systems that are run constantly achieve a 'steady state' where there is no appreciable change in environment or conditions in which components run, and thus the thermal loads created by powering down and then up etc don't exist. The supposition is that steady state components last a lot longer than thermally stressed components. That is certainly true, though bearing faces of hard drives and fans for example are prone to higher wear. In comparison, systems that are power-cycled have fewer 'component hours' and lower wear on bearing faces, though they do actually suffer torsional wear on bearings making them more likely to fail in variable speed situations.
All in all, it's really swings and roundabouts. In the days of relatively poor component quality and software bearings with high tolerances, it may have made a difference, but these days, nothing that can be reliably, or at least consistently, measured one way or another!
