Not on x86-64

I find it comical that you keep going on as if cdecl was the great plague or whatever - it has already been pointed out that the stack adjustment can be accumulated, resulting in smaller code than stdcall. And as bitrake points out, if you're not writing in a HLL you can do some nice stack re-usage tricks (never attempt this when interfacing with external code, though).

And really, just how much of a performance problem is this in the real world? I thiiiink there's more important things to spend your time on

If you're going to hand-optimize in assembly you can probably do better than any standard (including cdecl), so that's a moot point.

That's not the point. A wrong approach and poor "standard" is important to get right. No wonder things and standards (not calling conventions mind you) are so complicated these days, because everyone thinks "meh, this isn't really significant change, just a small one" but when you've got different people it will result in junk and the incompatibilities and all the problems we face today with so many incompatible devices/drivers/whatever.