Hi Phil,
You're running into a tradeoff we made in order to serve full downloads to customers at scale and make updates in the background so a download is always available.
The filtered files are produced by rolling over every feature that we have extracted from the underlying service and then stored on S3. There's no database here, just streams of data flowing to and from S3. Therefore the operation is the equivalent of a full table scan, plus network time to move the data back and forth from s3, plus the time it takes to convert the filtered data into the requested format. This clock starts after the job to create this file reaches the front of the queue.
Once the download starts you should receive the data with plenty of bandwidth.
I understand why you would want to receive a filtered download more quickly, but the current system is not optimized for that. With future work, we may be able to support faster filters.
I hope this helps you understand why things are the way they are. If not, I'm happy to try to explain further.
Daniel Fenton
Software Engineer | ArcGIS Hub