The model is set up to run forever - stop time is "inf". The two Error Rate Calculation (ERC) blocks are set up to stop simulation when 1000 errors are collected or 1e6 samples are processed, whichever happens first.
In the default set up (Eb/No = 0), the first ERC block collects 1000 errors in ~6600 samples and stops the simulation. When you increase the Eb/No value, the error rate reduces significantly. Therefore none of the ERC blocks hit 1000 errors even after processing 1e6 samples.
So, in the default case, the simulation stops after processing ~6600 samples, while in the later case the simulation continues until 1e6 samples are processed. Hence, the longer simulation time.