Sixty Seven Percent Rejected

Client-side adaptive throttling: each client tracks, over the last two minutes, requests (attempts it made at the application layer) and accepts (attempts the backend accepted). Before sending, it drops a request locally with probability

$p_{\text{reject}} = \max\left(0,\ \frac{\text{requests} - K \cdot \text{accepts}}{\text{requests} + 1}\right)$

A client's window reads requests = 3,000 and accepts = 900. The default multiplier is $K = 2$. The team then lowers $K$ to $1.1$ to protect the backend harder.

What are the two rejection probabilities, and what does the change buy?