Optical neural networks offer radically new avenues for ultrafast, energy-efficient hardware for machine learning and artificial intelligence. Reservoir Computing (RC), given its high performance and cheap training has attracted considerable attention for photonic neural network implementations, principally based on semiconductor lasers (SLs). Among SLs, Vertical Cavity Surface Emitting Lasers (VCSELs) possess unique attributes, e.g. high speed, low power, rich dynamics, reduced cost, ease to integrate in array architectures, making them valuable candidates for future photonic neural networks. This work provides a comprehensive analysis of a telecom-wavelength GHz-rate VCSEL RC system, revealing the impact of key system parameters on its performance across different processing tasks.