Hi Yolanda Guerrero-Rebollo,
Thanks fo watching this special, yes, I like to learn and to explore new things. As any other artists, I get bored quickly, if there is no progress. 
Thinking about 3600 frames with a total duration of three minutes would equal a frame rate of twenty frames per second. That might not look well. If the motion is slow, you might get away with it.
The render-time depends on many things. Given that I started roughly three decades ago rendering stuff out, a frame in 1K was pretty much a day back then. So three minutes sounds good, perhaps not. Studios who render frames for feature films can take days per frame, even with large render farms, while a feature film has around 120,000 frames. Again, it depends.
I think we could talk a day about improving a scene to get faster, only to notice that we have so much more.
I can say that baking as much information as possible is key to the texture or caching in animation. Besides that, taking objects out of the GI rendering calculation when they are not leading to significant changes but have tons of polygons, like plants, while replacing them (if needed) for the GI (not visible to the camera itself) as a low-polygon object.
It also depends on the render engine and how materials are set up. I have seen here in the forum (over many years) that expensive machines are bought, and then people threw just anything in the scene and let the CPU/GPU figure out how to handle it. I think it is clear how that works.
In Redshift, for example, one can fine-tune so many areas that this might take a while to go through all of them. Which should be handled in an Ask The Trainer “Render Special” 
Using render instances, for example, might help, or replacing things in the background with a 360ºx180º panorama, if the camera move allows for it (parallax).
Editing a movie is often more efficient, and a one-take with three minutes needs an audience who really wants to see it. Edits in typical movies are 2-8 seconds (+/-)long, and then the following clip forces the brain to interpolate what is missing based on the cut. Long ones taken are often not the best way, but I don’t know what you have set up. If the audience gains a lot from it, you have them on the edge of their seat for three minutes.
Can you explore the project by shutting off parts, checking how long 30 frames need to render, and exploring other parts? Perhaps there is no bottleneck, or there is something that slows it down. Again 3 minutes sounds not bad, but it depends.
Baking things into textures, switching parts off, and exploring the replacement of secondary (far away) parts are essential. This is typical work for each larger project, and it certainly is each time different.
Two decades ago, I had to deliver a 20-minute animation in one month, including starting from scratch, e.g., modeling a dozen scenes. I placed a lot of equirectangular (360ºx180º) of a single scene (360 camera rendering) as well as camera mapping into the movie, where the render-time (back then!) was not even 10 seconds per frame based on this “cheat”—speaking of editing and “baking”. Today those frames would render below a second.
In summary, your question provided very little to go on. It could be that motion blur or Bokeh takes the most time, perhaps to a certain degree, that can be done in post. (Not as good, but often nice enough).
Group your project in layers and switch some on or off and see if things change drastically.
My best wishes for your project
