I'm doing some tests with QField at the request of the company I work for.
I was wondering what is the best (most efficient and practical) raster format with regard to the fluidity and the loading speed while using the app on-the-field ?
The QField documentation recommends using GeoPackage raster for this purpose.
Other sources (citation needed) recommend mbtiles format.
Whatever you choose, here are a couple of things to consider:
Tiling tiled data sources allow for loading only parts of the data and not always reading the whole dataset while rendering.
Pyramids if pyramids are built, it's possible to render raster images much more efficient at small scale (zoomed out) because aggregated information is already available pre-calculated.
Compression to make files small enough to fit on a handheld device, consider turning on some sort of compression. Raster data, like any non-random imagery, can normally be compressed very well (jpeg, png, ...) and for visual purpose like you have here, there's not even a need for a loss-less compression.
Have a look at the above link that provide some hints how to prepare the data using gdal commands (many of them are also available as QGIS processing algorithms)