In this paper, we investigate a mobile edge computing (MEC) network, where the user has some computational tasks to be assisted by multiple computational access points (CAPs) through offloading. We consider practical communication scenarios with limited spectrum resources, and the cochannel interference arising from the aggressive reuse of frequency severely degrades the system offloading performance. To enhance the system performance, we provide three CAP selection criteria to choose one best CAP among multiple ones. Specifically, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">criterion I</i> maximizes the computational capability at the CAP, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">criterion II</i> minimizes the interfering power, while <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">criterion III</i> maximizes the instantaneous channel gain of data link. In time-varying channel environments, the CAP selection may be outdated, which deteriorates the system performance. For the three criteria, we evaluate the system outage probability in the outdated channel state information (CSI) by taking into account the latency, energy consumption and data rate, and provide the analytical and asymptotic expressions of outage probability, from which we obtain some critical insights on the system design. Simulation results are finally demonstrated to verify the proposed studies. In particular, criterion III under the perfect CSI can achieve the system whole diversity order coming from multiple CAPs.