It's definitley OK to do this; in fact we've included a number of features in v2 to help you target jobs to specific types of nodes should you need to schedule based on these hardware differences.
As you say though, it can cause some performance problems (especially with MPI jobs).
Thanks,
Josh