Is there one large Matrix or is there one for each crop?

--Pro--
One for Each Crop

Computer Power. The amount of extra computation required for each human that enters The Matrix most likely can't be illustrated in a linear equation (more likely an exponential formula of some sort). For each person, there is a huge array of sensory perceptions to calculate, and the more humans you have, the more they'll be affecting each other, creating even more computations. Note that I'm talking about computational speed, not electrical power.

Size. The more human beings put into the world, the more complicated and large their society is going to become. By limiting a user's possible areas of movement, you save memory and can maintain a much more manageable scope. The smaller area also ensures that hard to calculate variables, such as the weather of other areas of the Earth, don't have to be taken into account for the sake of one human being wondering in the rain forest after a plane crash.

Security. In the event that someone like Neo comes along, you can isolate the enlightened people to just one crop. This way the truths about the fabric of the matrix and so called "reality" doesn't spread, and you can simply wipe the memories of a small population. Also, in the event of a human disease, (perhaps spread by eating each other's ground up fluids) you don't have to worry about a massive power shortage because of great loss of human life.

--Con--
One Matrix

Bug Testing. As Blizzard found out from Diablo II, testing software with a small group of people can't provide accurate insight to the problems that will be encountered when you grant access to a huge number of people. There could be dangerous power hazards waiting to happen, that could only be seen through an immediate surge of users for example. It could also be hard to detect minor memory leaks that could be draining the efficiency of The Matrix.

Scalability. When developing a program it is much easier to scale a large program to a smaller size by stripping away code than it is to build up an infrastructure to scale up a program. For emergency purposes, large or small matrices could need to be created quickly, with little time for lots of software rewriting.

Time Management. Sooner or later, The Matrix is going to have to reboot. If time continued forever, then humans would eventually recreate AI inside the Matrix, and perhaps eventually a matrix within a matrix. Therefore you need to reset time to a preset date every so many years in order to prepare for the next generation. If multiple matrices are used, then they'll all be running on time based on when they were constructed, creating a large number of different time periods to manage. Having one center of operations ensures synchronization, and less data to sift through.

In the movie, Agent Smith says: "Did you know that the First Matrix was designed to be a perfect Human world where none suffered, where everyone would be happy? It was a disaster. No one would accept the program. Entire crops were lost."

The machines recreated the Matrix and modeled it after the peak of human civilization. One would think that they would try to maintain as many of the complexities of the real world to keep the virtual prisoners distracted from their true condition.

Also, the problem of computing power (speed) can be reduced from a purely quadratic growth function to a more linear one. Suppose that there is a limit to human perception. One can't percieve the breath of the person sleeping in another room. (Unless, perhaps, they are a loud snorer.)

The Matrix could draw "sandboxes" around groups of people sharing an experience. People isolated enough from eachother wouldn't have to share immediate experiences -- only when one entered into the other's "sanndbox". In this way, the Matrix can create many "sandboxes" and use a type of SMP to process separate groups' experiences.