If residuals of observed values farther from your prediction line have greater variance than residuals with observed values closer to your prediction line, then outliers are harder to detect.