Which is like saying it's pointless to buy a hammer when your screwdriver is 'heavy enough' to hit things. Honestly, it seems like you're hell bent on solving a non-existent problem with the wrong tools.
Sure, in theory the ability to 'remotely dial out' from a machine on one point of an IP network to a modem physically located within the local dialing area of the destination seems like a reasonable idea. But it starts falling apart when you consider the clusterfuck of complications involved with keeping modems working reliably. Ask anyone that's run a modem pool to talk about the nightmares. And re-using soundcards, while technically "possible" just makes it even more of a clusterfuck by adding a layer of software on top of what's likely to be a very unstable layer of audio hardware interfaces. There are just too many soundcards and too many variations to even begin to make it worthwhile even if there WAS a justifiable reason for this hack.
This contrasted with the idea that if something needs connectivity it will have to rise to the level of IP connectivity to make it happen. Sort of like riding a roller coaster, if you want the thrill you must be "this tall to ride this amusement".
While remote data collection devices with simple modem chipsets might be one application that could use them this idea there would still have to be POTS lines made available to them. Actually, what you're sort of talking about is almost a throwback to using X.25 networking clouds.
Anyway, there's a dozen different ways to skin the cat, I'm just not seeing your idea as being worth the hassles involved. It could be lack of vision on my part but given my two decades of being in this business I somehow doubt it.