We derive new upper bounds on outage probability (OP) and spectral efficiency (SE) for a simultaneous wireless information and energy transfer system under spatial correlation and optimal phase configuration at intelligent reflecting surface (IRS) when users are served based on round-robin (RR) scheduling, share common source to IRS links and adopt nonlinear energy harvesting. Diversity order for this system is characterized. We then extend our study to a multi-antenna source and analyze OP and SE under random and equal phase shift configurations at IRS. We design beamformers at the source and at IRS under different strategies, namely RR scheduling and simultaneous service with and without signal-to-interference-plus-noise ratio (SINR) constraint. Numerical results are presented to validate the accuracy of our statistical modeling and mathematical analysis and quantify the gain in performance relative to random and equal phase shifts. We illustrate that higher number of users can be served by increasing number of IRS elements while keeping OP fixed. We identify the operational regime where RR scheduling yields better performance than serving users simultaneously without SINR constraint. We show that increasing IRS elements can help maintain target harvested power even under stricter SINR constraint. Impact of estimation error on performance is illustrated.