Please look here, even though I think I should be more straightforward. The Dilate-Erode from Volume is unavailable in Fog, so I used the Thicken Object. The more I place into it, the more it feels like a hack, making me think about starting over. But perhaps that works for you. Let me know
I have placed a Copy of the original and set it to pale blue and X-Ray for comparison.
Speed set the Voxel size for the time of exploration as large as possible. Keep in mind that between 10 cm and 0.1 cm, the difference in amount of voxels is around 100^3 or 1,000,000. This is super simplified, but keep the values for the size large (Low amount!) and go only smaller (Higher!) when needed.
Perhaps download Cinebench and google for results to see how your setup works comparatively. An activity Monitor or Task Manager allows for some exploration if the system reaches a limit.
Switch off all apps and functions that eat into your performance. Google is typically one app that asks for a lot of performance. Enabling redshift only when needed is a current circulating tip.
Please remember that tutorials can be recorded and sped up, then the commentary is placed on top. The tutorial should be slower, as the screen recording is active at the same time.
I do not want to go deeper here, as this forum is not about hardware. But perhaps I have to suggest at one point to have an Ask The Trainer show about performance optimizations of hardware stuff.
However, as a friend told me three decades ago when he got his first NeXT system: Whenever something gets faster, we put more options in our software and have it a little bit slowed down in return. I feel that is not so much different these days. Of course, the primary trend is that everything gets faster (Silicon and code), and expectations might run hot at one point.
Your question was general, and with that, it is often complicated to find a one-size-fits-all solution, s we all have our ideas of where it would stop working.
If you have a specific case, then please share that as a sketch or as a file.
My typical MO is that I love seeing the Project file or a sketch.
I do not like to answer questions or reply from memory. Over the past 17 years, running the Q&A forum here has proven to be the best way. This thread shows that not having specific data is wasting your time. Sorry about this.
I love to find creative solutions if there is no default option.
One question, why it has to be a Spline? What is the target? The more details you share, the better I can explore this.
The best option is to either bake the information or cache it. If information is cached, it is available at the beginning of the frame. If information is not processed when needed, the available information is used, which means from the previous frame. This can add up in a change reaction.
On the other hand, caching and baking means that most of the work is done, and data is provided much faster.
Then there is in some cases that the machine is too slow, but the setting All Frames is chosen, so it can't play in "real-time". All frames are the little Icons in the middle of the timeline, with the A in it.
The easiest way is to cache first what other parts need, then the next, and so on. Sometimes the cache stores the information, but it requires shutting off the initial influence, for example, the Skin Object, to avoid double influence.