In this paper the problem of dimensioning dynamic optical WDM networks under multi-hour traffic and soft QoS constraints is studied. Unlike previous works, traffic matrices are generated according to a lognormal distribution, a more realistic model than the usual uniform approach. Additionally, instead of requiring hard QoS constraints (e.g. blocking probability under a given threshold 100% of time), soft QoS constraints that allow blocking to increase during some periods of time are considered. In multi-hour traffic, such soft QoS constraints seem to be a more natural way of dealing with customer SLAs. Results show that using lognormal traffic matrices and soft QoS constraints leads to a decrease in the number of wavelength requirements. In the studied cases, this decrease was about 10%.