Probably they just split the difference between bending the limit on the maximum and the optimal load. The 18pF value is also very common, so maybe they had it in stock.
The likely consequences of exceeding the datasheet maximum for the MCU by a small amount is that some small percentage of units may fail to start oscillating at temperature extremes (where the gm of the amplifier is low). Since this is just an evaluation board, not a commercial product, nobody really cares that much. Using a different value from the optimal load for the crystal will cause a slight systemic error in the oscillation frequency. Again, probably not all that important.
Of course the crystal oscillation frequency will change with temperature, but the amount of change should be in line with the specifications of the crystal. A possible non-temperature-related drift can occur from exceeding the maximum drive power of the crystal, which can be rather small for some types of crystal, but in this case is specified to be very reasonable (less than 50uW).
I must say that the specifications for the oscillator section of that chip are quite excellent in terms of how complete they are. Usually the specs are vague and seem more aimed at deflecting blame. Who is the manufacturer of that particular ARM chip? Looks a bit like Freescale's datasheet format.