In high-speed railway communication systems, the distributed antenna systems are usually employed to mitigate frequent handover and enhance the signal-to-noise ratio to receivers. In this case, jointly optimizing downlink power allocation with antenna selection (PAWAS) can enhance system energy efficiency, while the channel state and traffic density are taken into account. Besides, two typical kinds of terrains with sparse and rich scatterings and three traffic patterns including delay-sensitive, -insensitive, and hybrid traffics are investigated in this paper. We show that severe small-scale fading decreases the ergodic capacity, which is quantitatively analyzed, and proved to be proportional to the number of selected transmit antennas. In addition, in case of delay-sensitive traffic, we show that the PAWAS can be viewed as generalized channel-inversion associated with transmit antenna selection. In case of delay-insensitive traffic, we show that when multiple antennas are selected, the power allocation can be viewed as channel-inversion, whereas when single antenna is selected, it is traditional waterfilling. In case of hybrid traffic, we prove that the optimal PAWAS method can be given by separately solving the PAWAS of its delay-sensitive and -insensitive parts. Simulation results validate our theoretical results and demonstrate that proposed PAWAS can minimize the average transmit power in cases of arbitrary traffic density and channel states.