We compare new HERA data for the longitudinal structure function F_{L} with the predictions of different variants of the dipole model. In particular, we show that the ratio F_{L} / F_{2} is well described by the dipole models and is rather insensitive to the details of the fit. Fits to F_{2} are performed with the help of geometrical scaling (GS). Using the property of GS, we derive the bounds for F_{L} / F_{2} both for the different versions of the dipole model and in the general case. Finally, we briefly discuss how the higher Fock components of the photon wave function may affect these bounds.