In this case, it might be that the opposite is true, Daryl.
Bob, like pointed out on pages 7 and 8 of MIR Pro's preliminary manual, there are actually two numbers to take into account here - namely the host buffer size and MIR latency setting. The latter should either be 0 (which is usually ideal) or a number greater than the host buffer size. MIR latency below or equal to the host buffer size will give sub-par results.
IOW: I would try to leave the latency of the RTAS engine at 512 samples first, while resetting MIR Pro's latency to the default 0 samples. If this doesn't do the trick, I suggest to use a higher latency for the RTAS engine first (1024). As a last resort, it might be advisable to use at least twice the size of the RTAS latency for MIR Pro.
HTH,
/Dietz - Vienna Symphonic Library