I think the crucial difference here is the number of entities I have to trust. People can generally do a bit of research on a bank before they join one, if a bank really sucks you can switch, and if a bank seriously defrauds you then you know whodunit and can pursue legal recourse. By contrast if I'm using a supposedly dumb terminal provided by some random cafe then there's absolutely no way to prevent them from seeing everything I do [1], or from inserting arbitrary commands into any service I use. And if I use the provided computers at many locations, as you suggest, there's probably no way for me to figure out who it was that screwed me.
[1] You say "cryptography". I say "Analog hole".