That ignores the back ends to the back end. E.g. an IRS system that might be able to handle a million people with query/response transactions would almost certainly need different strategies to scale up 100 times.

>That ignores the back ends to the back end. E.g. an IRS system

Just make sure your component won't fail. Perhaps offer other departments help with sharding, be that consulting or paying for servers. Give a friendly error message that the other system has gone down, but your data is saved and will be resumed automatically upon next visit. Maybe retry later and email people when it succeeds.

>an IRS system that might be able to handle a million people with query/response transactions would almost certainly need different strategies to scale up 100 times.

I don't see why it would. Sort by name or SSN and do a roughly-even split across your servers.