Why when comparing two means (with t-test) the equality of variances is also important?

If you use the F-test to compare variances, then if you reject the assumption that σ1² = σ2², you will use one formula, while if you don't reject the assumption, you will use a different formula.