A Concurrent Programming Framework for Quantitative Analysis of Efficiency Issues When Serving Multiple Long-Context Requests Under Limited GPU High-Bandwidth Memory (HBM) Regime