If you will perform git clone/git fetch/git push over SSH, then java process will take amount of CPU usage (~0.5 cpucore per request depending on network payload).

To reduce this load on CPU, you may to rename repositories on disk to it natural naming in form <PROJECT_KEY>/<repo_name>.git, not a repositories/<some_number>.

After it, system administrator may create symbolic link to this directory for user git to enable native SSH protocol to be used for access into repository.

The SSH keys from web interace simple must be copied into /home/git/.ssh/authorized_keys file. Every time then users added it.

To track changes from repositories every one bare must have pre-receve or post-receieve hook enabled with one send updates to Stash via REST or something other posible way.

After this will be done, the load of CPU must be drammatically reduced on network traffic from/to Stash via SSH protocol.

Offtopic. It very hard now to find a repository on disk to see whats wrong within it. Cos Stash even not added git config name, or something else identity to this bares.