Assume I have a surface code with distance $d$ and an i.i.d error model with both single qubit depolarization and measurement errors, both with probability $p$.

In this case, one usually repeats the syndrome measurement cycles $d$ times to get the threshold. But what happens if one repeats the syndrome measurement cycles for $m*d$ times, with $2\le m$?

When I simulated this using stim, I found that the threshold decreases as $m$ increases, and the logical error rate increases with $m$. Is this a known phenomenon, and why this happens? Or maybe I have a problem with my simulations? Where can I find a discussion of this topic in the literature?