The issue is the compression. There's hundreds of individual assets, the process to compress or more accurately, uncompress the assets for use takes processor resources. Usually it only really needs to be done a few times when the game starts, when it loads the assets required. Basically when you get to a loading screen, the game is unpacking the relevant assets from those dataset files. Every time the game opens one of those datasets, it takes time to create the connection to the dataset file on the host system, then unpack the index of the dataset, and finally go and retrieve the assets needed.

Two things about this process: first, securing access to the file and getting the index is a fairly slow process. Allocating anything takes significant time (relative to the other steps in the process) and accomplishes nothing except preparing to load the relevant assets. It's basically just wasted time. The second thing is that compressed files are most efficient in making the total size smaller when there's more data in the file.

Very basically, the most simple compression, zip (aka "compressed folders" in Windows) basically looks through the files for repeating sections of data, it then replaces all that repeated content with a reference to the original data. The reference is much smaller than the data it replaces. This can also be referred to as de-duplication. In this way if you had a set of files that all contained mostly the same data, say text files with most of the same repeating messages, the resulting compression would be very high (smaller size) and this method is used for things like log files since there are many repeating dates, times, and messages with a few unique variances from line to line. This is an extremely basic concept of one style of compression that's very common, and certainly not the only way, and also not necessarily the method being used, or the only method being used.

If there's less content per compressed dataset file, there's going to be fewer opportunities for the compression to optimize the content to be smaller, so large similar datasets are preferable over smaller ones containing more diverse data.

This, combined with the relatively long open times per file means that programmers will want as few datasets as possible to keep the system from needing to open many files to retrieve the required data during load times, and to boost the efficiency of those compressed files to optimal levels.

If, for example, many smaller files were used, then yes, updates would be smaller. However, loading times could end up being doubled or tripled from their current timing. Given that you would, in theory, be leading data many times over (every time you load into a game or a map or something), compared to how frequently you perform updates, the right choice is to have updates take longer with more data required for download, so when you get into the game, your intra-session loads may be much faster.

With the integration of solid state storage in most modern systems, loading times have also been dramatically reduced due to the sheer speed at which files can be locked, opened, and data streamed out of them into working memory, but it's still a trade-off that needs to be taken into account. This is especially true when considering releases on PC, since PC's can have wildly different hardware and not everyone is using SSDs, or similar (fast) flash storage; perhaps on older systems or if the end user simply prefers the less expensive space available from spinning platter hard disks.

All of this must be counter balanced to provide the best possible experience for the end user and I assure you that all aspects of this process are heavily scrutinized by the people who designed the game. Often, these decisions are made early on so that the rest of the loading system can be designed around these concepts consistently, and it doesn't need to be reworked part way through the lifecycle of the game. It's very likely that even as systems and standards change, the loading system in the game will not, so if the game was designed with optimizations for hard disks (not SSDs) in mind, then that will not change until at least the next major release in that games franchise.

What isn't really excusable is when the next game from a franchise has a large overhaul, and the loading system (with all of its obsolete optimizations) is used for more modern titles; which is something I'm certain happens with most AAA studios. They reuse a lot of the existing systems and code to reduce how much work is required to go from concept to release, and hopefully shorten the duration of time (and the amount of effort required) to get to launch. Such systems should be under scrutiny at all times whenever possible, to further streamline the process and optimize it for the majority of players. If that means outlier customers trying to play the latest game on their WD green spinning disk have a worse time because they haven't purchased an SSD, when more than 90% + have at least a SATA SSD, all of whom get the benefits from the newer load system while obsolete users are detrimented because of their slow platter drives, then so be it.

But I'm starting to cross over into my opinions on it a bit more than I intended to. So I'll stop there. I hope that helps at least make sense of what's happening and why such decisions are made. As always if anyone reads this and knows more than I do, please speak up and correct me. I'm just some guy on the internet, and I'm not perfect. I don't make games, I'm not a developer. I am a systems administrator, so I see these issues constantly; I know how the subsystems work and I have a deep understanding of the underlying technology, but I haven't done any serious coding work for a long long time. I may be wrong or inaccurate on a few points and I welcome any corrections that anyone may have that they can share.

Have a good day.