Years ago I worked on a project that used files in the TIFF image format as databases. The TIFF format supports the concept of multiple files embedded within the main file, and multiple sections within an image file, each of which is indexed and can be accessed and compressed individually, with different compression methods.

TIFF was designed to be an image format, but of course nothing prevents you from storing and accessing arbitrary data.

In our project, TIFF files proved to work well as read-only databases of large collections of structured binary data.

I have no use for this since my data (.png files) isn't compressible but this is brilliant and I wish I had a use case for it!

Would it be possible with something even more space efficient like 7zip / LZMA2 ?

What about false positives when looking for the next header?