I'm not sure "proof" is the proper concept here since the variance of the t distribution is defined as the second moment of the pdf as for all distributions. The variance is not defined for n=1,2. For values of n>2 the variance is derived from Wallis integrals discussed here in Tutorial 2: