Atlas PR 17 <https://github.com/nim-lang/atlas/pull/178> modifies Atlas to use Git's remotes for URL management including handling forks and unofficial packages. This change means Atlas wouldn't need to use URL "triplet" folder names and cloning forks should be much faster.
See here for discussion: https://github.com/nim-lang/atlas/issues/179
It shouldn't really affect performance. I haven't noticed any slowdowns in my testing locally.
Forks are treated as extra packages internally and track the remote's name so that calls to git are given the remote to operate on for finding versions / tags / etc. There's some minimal overhead in setting the remote origin when setting commits.
What I'd like to next for speedup and to clean up things would be lazy cloning historical packages. Currently Atlas clones all possible repos in the history and then runs SAT. This mainly was to get SAT working initially.
However as discussed on recent threads most dependencies just use the latest version or head. Atlas already does lazy cloning for "features" so it should be fairly easy to add. The expandGraph core logic runs an iterative process and just needs to mark historical packages as lazy clones. Then run SAT and see if they're needed. Since the solver logic is separate from loading packages it's pretty straightforward.
Beyond that just adding caching or threading would probably get Atlas < 1-2s runs in most cases.