If you consider the propagator to be the following(check it for sign of metric):
$$D(x-y)=-i\int \frac{d^3k}{(2\pi)^32\omega_k}[e^{-i(\omega_k (x^0-y^0)-\vec{k}\cdot(\vec{x}-\vec{y}))}\theta(x^0-y^0)+e^{i(\omega_k (x^0-y^0)-\vec{k}\cdot(\vec{x}-\vec{y})}\theta(-(x^0-y^0)]$$
Then you are obviously right. The time-ordering of $x^0$ and $y^0$ is obviously not Lorentz invariant when the separation is space-like. It is clearly stated in many textbooks but let us provide just few references for those who are curios: see here-page 211 ,here-page 32, here-chapter4 or here-chapter10. I definitely recommend chapter 4 of Sakurai's old text.
But now let's get back to your question. We know that sign of time coordinate doesn't change under Lorentz transformation. So if we have $x^0>0$ and $y^0>0$ in one frame it will remain so. But the sign of their difference is not invariant, indicating that it can change when separation is space-like(as we said in above paragraphs). Now suppose that you take $y$ in the above formula to be zero. What happens after a Lorentz transformation? It is simple: zero remains zero, and if $x^0$ was positive or negative it will remain so. The result is that in this special case $sign{(x^0)}$ is Lorentz invariant.
Indeed the statement that $sign(x^0-y^0)$ is Lorentz invariant is a silly statement that you will never find in any book.