The performance of irregular phased array architectures is assessed in the context of multi-user multiple-input multiple-output (MU-MIMO) communications operating beyond 100 GHz. Realizing half-wavelength spaced planar phased arrays is challenging due to wavelength-integrated circuit (IC) size conflict at those frequencies where the antenna dimensions are comparable to IC size. Therefore, irregular array architectures such as thinned and clustered arrays are developed to mitigate the wavelength-IC size conflict. In the thinned arrays, radiating elements are permanently deactivated, while in clustered arrays, neighboring elements are grouped into subarrays. Furthermore, irregular arrays are integrated with hybrid beamforming architectures to manage the complexity introduced by full digital beamforming, where a single radio frequency chain is connected per power amplifier. An optimization problem is formulated to determine the optimal arrangement of antenna elements where the trade-off between spectral efficiency (SE) and sidelobe levels (SLL) can be tuned. Clustered array configurations are optimized by genetic algorithm and Algorithm-X based methodologies, where the former relies on a randomized search and the latter exploits brute-force search, respectively. Furthermore, a prototype array is designed on a printed circuit board (PCB) to verify the proposed methodologies through full-wave simulations. To have a fair comparison, clustered arrays with a grouping of two and four elements are compared with thinned arrays with half and quarter thinning ratios, respectively. The combination of hybrid and irregular array architectures leads to minimal or no performance degradation in the case of hybrid fully connected architectures but severe SE and SLL degradation in the case of hybrid partially connected architectures, respectively.