Depends on the case. But in general I always try to have these rules:
Positive pressure for the reasons already listed.
Fans exhausting across all spinning disks. I don't like to rely on passive flow over disk drives, so I always make sure to have a case that I can mount a fan directly in front of any spinning disks.
Front-to-back airflow, like
GodATX intended.
No major intake/exhaust points without a fan. If a case has 3 fan openings in it, I either fill all of them with fans or I seal the extra openings off. This helps prevent dust, and also helps me directly control airflow. This also goes for large vents in the case (like
these side vents on the CM Elite 110), which I seal off. This goes hand-in-hand with my next point.
Redundant fan layout. I try to set it up so that if any one fan fails, the passive air flow from the remaining fans will ensure there is no stagnant air. In smaller cases like the Elite 110 linked above, I can rely on the combination of the main fan in the front and the power supply's fan so that if the front fan fails, the negative draw from the PSU fan, combined with the fact that all the extra vents are sealed off means air will continue to flow front-to-back, across the hard drives. In a larger case, I try to make the setup close to neutral pressure, but slightly positive, so that if one intake fan fails, the case will switch to negative pressure, and the opening the failed fan is covering will still get a small amount of airflow. I try to also use at least one temperature-sensitive fan in the case so that any failure of the other fans will be somewhat corrected for by the temperature controlled fan.
I cut my teeth building machines in the height of the 150+w Prescott CPU era, where a single fan failure would get the system so hot it would practically turn normal CPU grease into epoxy (I've removed many heat sinks from boards with the CPU still attached, and had to pull the CPU from the heat sink with a vice). Many of my builds are for machines in a factory environment without HVAC so I tend to make sure I over-think cooling.
The part that always bugs me is I still don't think there's a proper standard that takes into account the massive wattage output by modern GPUs. I greatly dislike the unorganized nature of the airflow created by GPU fans that just recirculate into the case, but the tiny ~10 cm
2 opening on the second PCIe slot cover really isn't a substitute for even the ~100cm
2 opening on a 80mm fan, which is the smallest fan usually on a CPU heatsink, even though the CPU consumes ~90w and the GPU can be well over 200w.