First, there's a lot that can go wrong with a parachute + air bag landing system. Second, the Mars Pathfinder / Mars Exploration Rover systems also required rockets to fire just before reaching the ground. In that case they were solid rocket motors as compared to the throttled liquid monopropellant engines used by Mars Science Laboratory. Third, all of the systems use a parachute.
On the first point, an air bag landing system is vulnerable to ripping open the bags on impact against a natural surface with rocks. Or even without rocks if you hit hard enough. The solid rockets fired at terminal descent only provide coarse control over the impact velocity, which is why the air bags were required. It was an approach to develop a low-cost Mars landing system that did not require more expensive throttled thrusters or a more capable landing radar which would be needed for finer control over the impact velocity. While lower cost, the air bags could be taken out by a very rough surface, high horizontal velocities due to wind, or a high total velocity of impact due to radar spoofing from, for example, varying terrain such as mesas.
When first proposed, the air bag system was not viewed as robust, but rather extremely risky due to the velocity of impact, the number of impacts (let's replace one landing with thirty!), and the associated accelerations. And it just plain looked crazy. That risk was accepted in order to lower cost.
In the end the system did work, with some changes in MER over MPF to improve the reliability by compensating for some wind effects and significantly toughening the air bags. The funny thing is how before it flies everyone says it's crazy and will never work. After it flies and works, and you try to go do something else, now the new thing is crazy and everyone asks why you're not using that really robust air bag system? Go figure.
As noted, an air bag landing system does not scale well when increasing the rover from 170 kg to 900 kg. The only way to get an air bag system to work at that scale would be to reduce the landing velocity. So you take the rover and air bags hanging on the tether in MER, and replace the solid rockets at the top of the tether with throttled hydrazine thrusters and replace the altimeter radar with a doppler radar for better control of the impact velocity. Once you take those steps, you can reduce the landing velocity dramatically. So much so, that you can eliminate the air bags entirely! You can even eliminate the lander structure which avoids having to get the rover off of a lander to start the mission. You need to toughen the wheels and suspension to have them serve as landing gear, but that is a happy price to pay to get rid of all that other stuff. Voila. You have the skycrane.
If you want to make a landing system as reliable as possible, you reduce the impact velocity as much as possible. The complexity required to lower the velocity does not inherently make something unreliable. It just means it will cost you more to make it reliable. In return, you can make the landing much less sensitive to the atmospheric and surface environments, which you have no control over. Overall, the MSL landing system is much more reliable.
