Measurement of the penetration depth lambda(T) as a function of temperature using millimeter wave transmission in the range 130-500 GHz are reported for three YBa2Cu3O7-delta (YBCO) laser ablated thin films. Two films, deposited on a LaAlO3 substrate (T-c = 90.2 K), exhibit a narrow resistive transition (0.3 K). One has been subsequently irradiated with He+ ions in order to increase the scattering rate of the quasi-particles (T-c = 87.8 K). The third film, grown on MgO (T-c = 88.5 K), exhibits also a fairly narrow transition (0.8 K) and a high crystalline quality. The experiment provides the absolute value lambda(T less than or equal to 30 K) for the penetration depth at low temperature: the derivation from the transmission data and the experimental uncertainty are discussed. We find a zero temperature penetration depth <(lambda(0))over tilde> = 1990 +/- 200 Angstrom, 2180 +/- 200 Angstrom and 2180 +/- 200 Angstrom, for YBCO-500 Angstrom/LaAlO3 (pristine), YBCO-1300 Angstrom/MgO and YBCO-500 Angstrom/LaAlO3 (irradiated) respectively. lambda(T less than or equal to 30 K) exhibits a different behavior for the three films. In the pristine sample, lambda(T less than or equal to 30 K) shows a clear temperature and frequency dependence, namely the temperature dependence is consistent with a linear variation, whose slope decreases with frequency: this is considered as an evidence for the scattering rate being of the order of the measuring frequency. A two fluids analysis yields 1/tau (T less than or equal to 30 K) similar to 1.7 x 10(12) s(-1). In the two other samples, lambda(T less than or equal to 30 K) does not display any frequency dependence, suggesting a significantly larger scattering rate. The temperature dependence is different in these latter samples. It is consistent with a linear variation for the YBCO/MgO sample, not for the YBCO/LaAlO3 irradiated one, which exhibits a T-2 dependence up to 40 K. Mie have compared our data to the predictions of the d-wave model incorporating resonant scattering and we do not find a satisfactory agreement. However, the large value of <(lambda(0))over tilde> in the pristine sample is a puzzle and sheds some doubt on a straightforward comparison with the theory of data from thin films, if considered as dirty d-wave superconductors.