What happens to your files
Both tabs run on JSZip 3.10.1, a JavaScript ZIP library loaded from the cdnjs content delivery network, with FileSaver.js 2.0.5 to hand the result to your browser's download prompt. Once the page and those two scripts have loaded, nothing further goes over the network: the files you drop in are read with the browser's File API, processed in memory, and the archive or extracted file is saved by your browser. Our server never receives them, which also means we cannot recover a file you forgot to download. The analytics script in the site header records the page visit, not your files.
The trade-off is memory. JSZip holds the whole archive in the browser's memory, so an archive of several gigabytes can make the tab slow or crash it; for anything that size a desktop archiver is the better choice. The library also cannot open password-protected or multi-volume (split) archives, and it reads ZIP64 archives only when they are small enough to fit, as its documented limitations explain.
Why some files barely shrink
DEFLATE finds repeated byte sequences and replaces later copies with short back-references, then codes the result with shorter bit patterns for the most common symbols. Text, CSV, JSON, HTML, source code, logs and uncompressed bitmaps are full of repetition and shrink a great deal; the whitespace in pretty-printed JSON, for instance, is exactly the kind of repetition DEFLATE removes, so a file you have tidied with the JSON formatter zips to nearly the same size as its minified version. JPEG, PNG, WebP, MP3, MP4, most PDFs and modern Office files (DOCX and XLSX are themselves ZIP archives) have already had their redundancy removed by their own formats, so zipping them saves a few percent at most and sometimes adds a little overhead. If a folder of photos barely shrinks, that is the format, not a fault. To make images smaller, re-encode them with the image compressor; zip them afterwards only to bundle them.
The ZIP format in brief
ZIP is specified in PKWARE's APPNOTE. Each file is stored independently with its own local header, compression method and CRC-32 checksum, and a central directory at the end of the archive indexes them all. That is why a reader can list the contents without decompressing anything, and why one damaged entry does not take the others with it. Because files are compressed one at a time, ZIP cannot exploit repetition between files the way the solid mode of 7z does, but any single file can be pulled out on its own, which is exactly what the Extract tab relies on. Encryption is part of the specification, but JSZip does not implement it, so this tool neither creates nor opens encrypted archives.