cells at 30-50% confluency are actively dividing cells. That means there is no contact inhibition which might interfere with your drug potency. In order to figure out the effect of your drug on the cell cycle, scientists use actively dividing cells. Some drugs induce cell cycle arrest at the G1 phase and others at the G2/M phase.
A low seeding density of 30-50 % allows for exponential growth of the cells during the time of treatment and allows you to observe anti-proliferative effects. It avoids reaching 100 % confluency, and thus contact inhibition during the time of treatment. If you start treating your cells at > 90 % confluency, you won't be able to detect a loss of proliferation compared to a control, because your control cells are inhibited by contact inhibition as well. This can lead to wrong conclusions.
However, depending on the assay, it can also be necessary to start with a confluent layer of cancer cells, e.g. in some cytotoxicity assays. Think about what you are going to analyze and design your experiment appropriately.
I agree with both, Hamadi and Carsten. Moreover, in order to test possible changes on cell proliferation induced by drugs or other agents, a low cell density when start treatment allows us to observe and quantify both, anti-proliferative and proliferative effects.