Hi -- that is peculiar, BAYAREALIKE+J is usually very similar to the other +J models. It's somewhat possible that the BAYAREALIKE model gets very different ML parameter estimates, and when these are used as the starting points for BAYAREALIKE+J, it takes longer to get into the right zone of parameter space.
You might consider running with the optimiser optimx instead of GenSA -- the latter is slower & more thorough, but probably overkill for the 2/3 parameter models.
My gut feeling is your BAYAREALIKE+J has got to be just about done. But if you decide to kill it, just survey the screen printout and write down the parameter values producing the best lnLs, you can use these as starting points in a new search...
Cheers,
Nick