n8n in my experience using it for 2 years now, and compared to similar solutions as there are many doing quite the same thing: it is just a very good product. It's stable. It has a gazillion integrations out of the box and is architectured as a module system so it's easy to create your own. It is very community centric, with community workflows and also integration modules people kindly publish on GitHub.
Oh, it's open (core) source. And while certain (just a couple of) enterprise features should have been made open to qualify as being called open source, it's very close to that. Most powerful features are open, ready to self host, modify and make your own.
Does it end up driving webs of python partials forming apps. Absolutely. Does it scale ? It does. Do complex flow remain maintainable? As a coder I prefer to maintain a repo of code than visual elements made of snippets. But, the critical advantage is productivity, for simple flows the community intelligence solves everything so you can get an operational set of valuable solutions within hours, even minutes once proficient with the interface. Another factor is, you can deploy pilot flow acting as applications, test them with production data, and make that live with the press of a button once pilot testing is done. With a code project you would need a robust and well polished cicd pipeline to get that.
The limit or cons to me is a logic and compute heavy solution just isn't a fit to run on an n8n platform, scaling n8n just isn't as intuitive as scaling pure application component that do one thing.
An example you may have a cpu heavy node, and a memory heavy node. It makes scaling the whole instance very inefficient. Scaling memory of a dedicated memory intensive application and scaling compute for the compute intensive component simply is far more optimal.
If resource cost is not significant relative to the value of your flows then just scale a self hosted n8n and you only need to digest having to maintain, following your analogy, a "nest of pythons".
Note: n8n sadly only supports python or JavaScript for custom code nodes, would have been nicer had they built a polyglot runtime instead. That's however more than what every other flow platforms let users do.