People have been pushing for overflow checks for integer arithmetic on C on the basis that the cost isn't too high. For floating point it can only be lower.

We are probably talking about single-digit percent increases in run time. For some software that's relevant, but not for most.