Would it be safe to assume then, that Google tests it's software upgrades with 0.02% of it's gmail account? (However many thousands clients are affected, it's probably better than affecting ALL their clients).

I wonder how they choose their 'test' client base.