Linux

In regular FHS distros, an upgrade to libxyz can be done without an update to its dependants a, b and c. The libxyz.so is updated in-place and newly run processes of a, b and c will use the new shared object code.

In Nix' model, changing a dependency in any way changes all of its dependants too. The package a that depends on libxyz 1.0.0 is treated as entirely different from the otherwise same package a that depends on libxyz 1.0.1 or libxyz 1.0.0 with a patch applied/new dependency/patch applied to the compiler/anything.

Nix encodes everything that could in any way influence a package's content into that package's "version". That's the hash in every Nix store path (i.e. /nix/store/5jlfqjgr34crcljr8r93kwg2rk5psj9a-bash-interactive-5.2-p15/bin/bash). The version number in the end is just there to inform humans of a path's contents; as far as Nix is concerned, it's just an arbitrary name string.

Therefore, any update to "core" dependencies requires a rebuild of all dependants. For very central core packages such as glibc, that means almost all packages in existence. Because those packages are "different" from the packages on your system without the update, you must download them all again and, because they have different hashes, they will be in separate paths in your Nix store.

This is what allows Nix to have parallel "installation" of any version of any package and roll back your entire config to a previous state because your entire system is treated as a "package" with the same semantics as described above.

Unless you have harsh data caps, extremely slow connections or are extremely tight on disk space, this isn't much of a concern though.
Additionally, you can always "garbage collect" old paths that are no longer referenced and Nix can deduplicate whole files that are 1:1 the same across the whole Nix store.