So there was a bug in my autotune-feature for the thread-configuration. Apparently somehow this didn't have a noticeable effect with the pre-fork-algo, but since the fork the resulting hashrate is much worse.
So after fixing the bug the hashrate for 10 instances of F2 is back to around 1.4kH-1.6kH, and it does not seem to be dependent on the region so much anymore.
In order to activate the fix for your miners you can either restart your pool (scale to 0, then scale back to e.g. 10 instances) or you just wait for ca. 3 days (after running for 3 days the miners will fetch the latest version of the script from github).
Your new script seem good but may be need some more tune to get better hash.
Can you try to add this to the script :
"sudo sysctl -w vm.nr_hugepages=128"
[2018-11-12 01:41:41] : Mining coin: monero
[2018-11-12 01:41:41] : Starting 2x thread, affinity: 0.
[2018-11-12 01:41:41] : hwloc: memory pinned
[2018-11-12 01:41:41] : Starting 2x thread, affinity: 1.
[2018-11-12 01:41:41] : MEMORY ALLOC FAILED: mmap failed, check attribute 'use_slow_memory' in 'config.txt'
[2018-11-12 01:41:41] : MEMORY ALLOC FAILED: mmap failed, check attribute 'use_slow_memory' in 'config.txt'
[2018-11-12 01:41:41] : hwloc: memory pinned
[2018-11-12 01:41:41] : MEMORY ALLOC FAILED: mmap failed, check attribute 'use_slow_memory' in 'config.txt'
[2018-11-12 01:41:41] : MEMORY ALLOC FAILED: mmap failed, check attribute 'use_slow_memory' in 'config.txt'[2018-11-12 01:41:41] : Fast-connecting to pool.supportxmr.com:5555 pool ...
[2018-11-12 01:41:41] : Fast-connecting to pool-ca.supportxmr.com:5555 pool ...