Why not just return a 429 (Too Many Requests) if the specific repo has been requested by google not too long ago ? (e.g. 1 hour earlier or).

It's a standard response code, and with a bit of luck google will scale the requests accordingly. If not, this will still allow the proxy to operate properly without burning too much server resources.

(I understand that this may leave some customers unhappy since the proxy may be up to 1 hour stale. If that's the case, the server could also add a check to see if there have been pushes since last request, though this is quite more complex)