If the assumption of homoscedasticity is violated (i.e., heteroscedasticity is present in errors) while all other CLRM assumptions hold, the OLS parameter estimators remain:

The unbiasedness and consistency of the OLS estimators depend only on the exogeneity condition E(u | X) = 0, which is not affected by heteroscedasticity. The estimators therefore remain unbiased and consistent. However, the Gauss-Markov theorem requires errors with equal variance. When this fails, OLS is no longer BLUE. It no longer has the minimum variance, and weighted least squares (a GLS procedure) gives more efficient estimates. There is a further consequence. The usual OLS formula for the standard errors is no longer valid, so the t and F tests become unreliable. The remedy is to use heteroscedasticity-robust (White) standard errors, or to apply WLS or GLS. Hence the correct option is (D).