improve worker request timing out??

Hi,
how do i improve p999 where it seems like worker times out?

This is a worker proxy for an R2 bucket. I use the bucket to store mp4s. i have code that makes lots of concurrent fetches for this media using remotion (distributed video rendering library)

sometimes i see:
Tried to download file "" but the server sent no data for 20 seconds

or just generic timeouts