The jitter of a free-running VCO with white noise sources is proportional to the square root of the time between the two observed transitions. See section 10 of http://www.designers-guide.org/Analysis/PLLnoise+jitter.pdf for details. So, it does not make much sense to talk about the peak-to-peak jitter of a free-running VCO because it is infinite if you observe it for an infinite time.

Based on what loose-electron suggested, I would wait for a "representative" number of cycles for guaranteeing 6 sigma distribution. I´m assuming the pk2pk jitter to be six sigma provided we see the cumulative Gaussian distribution.
Such number for me would be about 1000 cycles, since that should be the min # of cycles in order to see 1 event outside the six sigma population.

Rgds
Tosei