This is strange - if you are not making any changes to your model between runs, then the Accelerator target should not rebuild. Perhaps you should report this as a bug to MathWorks Tech Support.
Regarding your question about whether you might be better off with Normal Mode simulation - it depends. Yes, there is an initial overhead for generating and building code in Accelerator mode, but if your simulation is very long and expensive, then that initial overhead may be relatively cheap. The best way to figure out is to time your simulation. Using the tic-toc commands:
in both simulation modes and see which works out faster.
Best Answer