Why some torrent files are split into many small rar files? Does this improve the download speed? Or is it better for the trackers? What's the rationale behind this?
It is neither better for trackers nor does it improve download speed. One could easily argue that you can retain seeders in a torrent longer by maintaining the content in a directly usable format.
A lot of stuff that... uh... "pushes the boundaries of copyright law" is distributed initially by various groups on the alt.binaries newsgroups [EDIT: actually FTP topsites, see comments]. This is to save bandwidth, since the entire file only needs to be uploaded once, instead of to everyone who wants it until a seeder network is established. These get downloaded by a few people who then seed them to the torrent networks.
The multiple .rars are legacy from the usenet source, since many newsgroup servers have a maximum attachment limit -- as well as the aforementioned ability to easily download a replacement or use a parity file to recover damaged parts (things get damaged a lot in newsgroups). This doesn't matter in torrents, but it's a legacy of the initial source of whatever content it is.
It does not effect the download speed. The compression however decreases the time it takes to download as the file size is smaller. This could be done with a single rar file though.
There's a handful of motives for splitting into multiple parts.
- If a file becomes corrupted it's easier to download a replacement part (although modern torrent clients shouldn't have a problem even if it's a single large file)
- Back in the days of usenet, there was a limit on the size of attachments to each post, so the file was posted in many smaller parts. It was also common to protect your rars with "pars", which only worked with multiple small files.
The initial distribution of a 'release', is usually through various FTP servers at which the uploading group (or release group) is affiliated. Sometimes these members of the group live near one another and so when whoever created the initial release has finalised it they can split up the release as according to the scene rules for that type of release, and distribute it to their friends, so all their residential upload speeds can be combined to improve their chances of winning the race (being the first release group to upload that specific release).
But the splitting of file uploads is really a throwback from the time when we had very slow and unreliable Internet connections, thus connections breaking in the middle of an upload (or download) was not uncommon, meaning a large (then, 100+ MiB) transfer failing could be seriously inconvenient (and maybe cost you winning a 'race'). Hence rules were made to split releases into several smaller files (specifically chosen depending on overall release size), so that a failed transfer would mean only a small part of the release needed redownloading.
If you mean the actual .torrent files, they are (usually) so small that compressing them won't really matter.
For the files actually downloaded in the torrent client loading the torrent file, there isn't really any reason either if it's just one large file split into several rars. For several files split into rars, there could be a use if the user only wants to download the rars containing specific files (selecting only them for download).
As files in a torrent are downloaded in small pieces, further splitting these files by rars is a waste of time (assuming the files are not compressing well, like video files etc., for HUGE text files there could be value in it I guess, but not by splitting into several rars.).