This is a map of every census block in the USA, including its population. The entire file is over 12GB, but FlatGeobuf fetches only the tiny subset of data that intersects with the bounding box (drawn in blue). Pan the map to move the query's bounding box.
When you have feature data that cover a large area in fine-grained detail like this, the typical options are to either manually slice up your file into manageable regions or to rely on running an application server which does this slicing dynamically.
Hosting an application has initial complexity and ongoing maintenance costs. Slicing files can be tedious and inevitiably you might be interested in an area on the boundary of slices.
For these cases, consider instead using a single indexed FlatGeobuf. Because FlatGeobuf's spatial index allows you to fetch only the data you're interested in, you can keep your page size down while avoiding the tedium of slicing up files manually, or building and maintaining an application server.
Though getting a small subset of features from a large dataset transfers relatively little data, it does so using multiple requests. There are some tips for optimizing remote hosted flatgeobufs, mostly oriented around minimizing round trip latency.