What is the risk of performing the t-test using pooled variance, if the variances between the 2 samples are actually unequal?