I put domain names in /etc/hosts ks (static nginx on i5) cat-v.org.ovh (werc on rc-httpd on 9front with cwfs on kvm on i5) cat-v.org.rc-httpd (rc-httpd on rc-httpd on 9front with cwfs on kvm on i5) cat-v.org.ramnode $ ping cat-v.org PING cat-v.org (22.214.171.124) 56(84) bytes of data. 64 bytes from ks (126.96.36.199): icmp_req=1 ttl=60 time=9.23 ms 64 bytes from ks (188.8.131.52): icmp_req=2 ttl=60 time=9.23 ms time wget -O /dev/null -q http://cat-v.org ) 2>&1 | grep real p->ks real 0m0.034s real 0m0.039s real 0m0.034s real 0m0.034s real 0m0.034s real 0m0.033s p->cat-v.org.ovh real 0m0.170s real 0m0.163s real 0m0.159s real 0m0.169s real 0m0.159s real 0m0.162s p->cat-v.org.rc-httpd real 0m0.064s real 0m0.065s real 0m0.062s real 0m0.069s real 0m0.057s real 0m0.067s p->cat-v.org.ramnode real 0m0.431s real 0m0.436s real 0m0.432s real 0m0.427s real 0m0.438s real 0m0.432s real 0m0.438s run 10 times: time ( for bla in `seq 1 100`; do ( time wget -O /dev/null -q http://cat-v.org ) 2>&1 | grep real & done; wait) 100 requests started in parallel take about 10s on cat-v.org on ramnode and 11s on cat-v.org. on ovh, nginx only needs 430ms for the whole thing. With 400 requests to nginx in parallel this takes a bit longer: 1.8s If I cut werc out of the loop (only use rc-httpd) I can see the whole thing being done in 4s. If I up the number of parallel request to 200 I get 9s. Even if 9front wouldn't cache, the linux host that carries the disk images does it either way.