While doing clock domain conversions (rate matched) we usually double flop the data to avoid meta-stable states. Double flopping just reduces the probability of meta-stability. Triple flopping will reduce it further.
How to calculate the probability/relationship between meta-stability and number of clock domain flops used?
The canonical answer to metastability queries always involves referring to articles written by the late, great, Peter Alfke. In particular, the XAPP094 appnote - don't worry about the age of it, the theory is still the same.
There are also numbers for some more recent families available - although I can't see anything for the 6 and 7 series as yet.