In ob-C.el, line 339 has the format specifier

	    (`floatp '("double" "%f"))

to print literal floats. However, that format specifier rounds, which
can result in errors. I think the correct approach is to print the
float without rounding or padding or trying to guess at the desired
accuracy (and let the compiler do whatever is needed):

           (`floatp '("double" "%s"))

Attached is a patch with a test that illustrates the problem (it fails
on HEAD but passes with the patch).

Leo