Ah the good old days… Though technically speaking it wasn’t GenieACS, it was ‘genied’ which was its name initially before it was open sourced as GenieACS 
The biggest deployment that I’m aware of is just over 300k. With the right deployment architecture you could go well beyond that.
We’re getting there, albeit slowly. The new documentation hosted at docs.genieacs.com is at least pretending to look professional 
Not to hijack the thread but let me share a bit of context as to why simple things like this are more complex than they may seem:
At some point in the future I plan to make Genie be able to skip executing the scripts that it knows won’t have any effect (i.e. all declare statements are guaranteed to be already satisfied). This would have a significant performance improvement but Genie is already plenty fast so this is not really a priority at this point. For this kind of black magic to even be possible we need the execution context to be predictable. So any information leaking from outside of the CPE data model (e.g. HTTP headers) is going to be a problem in the future. That’s not to say it’s not doable, it just needs to be planned right.