"Life expectancy" can be interpreted a few ways, and I cannot find details on how NASA defines it. In humans, life expectancy is generally defined as an average age of death, which includes mortality from all causes at all ages. A space probe life expectancy calculated this way would include possibilities of the probe exploding on the launch pad or any other failures very early in the mission. So long as you avoid those very early failures (which are likely low probability and would not be well represented in a small sample size), your life expectancy now exceeds the overall life expectancy. Basically, a probe that makes it to Mars is likely to exceed its expected lifespan, since it has already survived through part of its life.
Another possibility is that this isn't a life expectancy in the traditional "mean age of death" sense, but a design expectancy. A design expectancy could have a much higher threshold of confidence for meeting a particular goal. It might not make much sense to plan a 90-day mission for Opportunity if the designers did not have high confidence that Opportunity would actually survive 90 days. In a traditional life expectancy calculation, you might find that roughly half the samples fail before the life expectancy time. I'm not convinced that NASA would launch a mission expecting a roughly 50% chance that the mission will fail before achieving its planned goals. A design expectancy could have a higher threshold, meaning that most missions should indeed survive to their design goal and beyond.
Ultimately, it might come down to exactly what is meant when discussing "life expectancy", and whether that's the traditional statistical usage (mean time to failure is 90 days) or a more colloquial/design usage (we expect 90 days of failure-free operation with high confidence). If it's intended to represent mean time of failure, it may indeed be an overestimate. If it's intended to represent a time before which failure is very unlikely, it may not be. I lean toward the latter interpretation - in a scenario with a long tail of survival times, more than 50% of samples will fail before the mean failure time. It would be rather poor PR for missions to fail to meet their expected lifespan more than half the time. I would interpret these as design expectancies, meaning it should be highly likely for missions to survive beyond the "expected" time.