Palm vein recognition has emerged as a promising biometric feature for large-scale identification, primarily due to its contactless nature, resistance to spoofing, and durability. However, current deep learning models face challenges such as overfitting, limited public datasets, and poor out-of-domain generalization. Recent studies suggest that a single CNN trained in a multitask fashion can improve data efficiency, reduce overfitting, and allow fast learning by leveraging auxiliary information. For this reason, we propose a multitask convolutional neural network (CNN) that integrates soft biometric features, specifically, gender and age, to improve palm vein-based individual identification. By leveraging shared representations across tasks, our model improves accuracy, efficiency, and scalability compared to existing single-task models. The proposed study introduces a new evaluation framework as a baseline for assessing and comparing CNN-based multitask approaches for palm vein recognition. Experiments on the VERA dataset demonstrate that the proposed model outperforms recent methods in identification accuracy while reducing computational cost. Furthermore, the results show a significant impact of weighted loss schemes, progressive warm-up training strategies, and adaptive learning rate strategies on the optimization of multitask models. These findings highlight our approach as a viable solution for deploying reliable and lightweight biometric systems in real-world applications.