The evolutionary cost of developing a particular trait is an attempt at quantifying how difficult the trait is to develop. Though the theory of evolution says organisms will tend to optimize fitness and do whatever it takes to survive, those predictions aren't the core of the theory and cannot be taken literally. Numerous times in nature a species will be confronted with a problem and never find a solution, instead going extinct; or they will take millions of years to exploit a resource which makes them prosperous. At other times these things happen quickly. Saying some adaptations have a higher cost than others is a way of quantifying or conceptualizing this difference.
Distance Within Fitness Landscape
Evolutionary biologists use the concept of a fitness landscape to picture how organisms evolve over time. Imagine every possible shape the organism could take on is represented as a point on a landscape. Evolving can be represented as movement on the landscape, and we can picture the organism slowly changing shape and behavior as it moves. Really, of course, there is variation within any species so the species is a small cluster of points on this map. Higher fitness is represented by higher altitude, and the points at higher altitude will reproduce more and lead to more points nearby. As a result the cluster as a whole will tend to climb hills.
In this model, evolutionary cost has to do with the distance along the fitness landscape. It takes longer to climb further. However, the current hilltop may not be the highest one around. To become optimal a species has to have offspring somewhere else than the current highest-known point; it must cross valleys in order to reach mountains. With random offspring this will sometimes occur, but not very often. So the evolutionary cost of a trait is higher when there is a valley between the current state and that trait.
According to the quasispecies model created by computational biologist Manfred Eigen, traits can be viewed as messages transmitted in the genetic code. Mutations are errors in these messages, and the only way a message can be maintained is if the errors are corrected. Death is the only source of correction in evolution. It turns out (according to Eigen's analysis) that the number of deaths required to maintain a trait is directly tied to the amount of information used in transmitting that trait. Essentially, a gene twice as long mutates about twice as often and so requires twice as many deaths to eliminate the mutations.
Yet obviously the total percentage of a species' population which can die per generation is limited. Thus this limit puts an upper bound on the amount of information a species can carry in its genome (no matter the actual length of the genome). In this view of evolution, a complex trait has a high evolutionary cost because it requires a large amount of information to implement it. Organisms developing complex traits must come closer to their information limits (thus causing more deaths per generation), or give up other complex traits in order to fit in the new one.
Of course, it is difficult to quantify precisely how much information a given trait takes, short of discovering the length of its genes. However, traits which take a lot of English to precisely describe probably contain a lot of information.