A manufacturer of relays will test hundreds of samples at higher and higher levels of random vibration. They will count the percentage of samples that failed at each level. They do statistics on the whole data set to estimate, with 95% confidence, the Grms where no relays will fail, and that value turned out to be 10 Grms. 

Let’s make up some data and imagine that at 30 Grms, 20% of relays failed and 80% of relays passed.  Now you have an analysis that predicts 30 Grms. Will you be satisfied with a 20% failure rate? I don’t think so. Perhaps the manufacturer only found 2% failed at 30 Grms. We don’t know, they don’t publish that level of detail.

All we know is that the datasheet says, don’t use these relays where the random vibration level exceeds 10 Grms. I suggest a redesign. Either change the PCB support or layout of where those relays are on the board (move them closer to the supports that see only 6 Grms) or choose a different relay that can survive 30 Grms.