The first laser in history used a synthetic ruby rod with mirrored ends. Subsequent ruby lasers also used this approach. This was apparently disregarded at an early stage in laser history, as I've been told that only very old laser rubies contained mirrored ends. Aside from permitting optional intracavity optics and the introduction of Brewster cut rods, there might be an additional reason for the apparent discontinuation of mirrored rubies.
The ruby rod stores energy. The energy can be released through spontaneous or stimulated emission. Spontaneous emission occurs after the 3 ms fluorescence time has elapsed, and it results in fluorescence wherein the light waves are uncorrelated, independent, and random in both time and direction. The bandwidth is also relatively broad.
Stimulated emission occurs when a population inversion is established, and amplification takes place when optical feedback is provided. Feedback of stimulated emission, through properly aligned mirrors, allows stored energy in the ruby to be released as coherent light. Without this optical feedback, spontaneous emission will eventually occur.
Stimulated emission will produce light waves that match the frequency, phase and direction of other light waves that have been emitted through stimulated emission. If the resonator mirrors are far enough apart, only the straightest light waves will be returned and 'cloned' through stimulated emission. If the mirrors are very close, some of the off axis stimulated emission might contribute to the output beam (especially in a high gain system like ruby, because relatively few reflected passes are required).
Imagine two parallel mirrors connected by a line that is perpendicular to their surfaces. Imagine that this line connects a point located at the center of each mirror. The smaller the distance between these two mirrors, the larger will be the maximum possible angle between this line and any additional lines that connect the two mirrors at any other two points. Put another way - the greater the distance between two parallel mirrors, the more parallel light must be in order to be reflected more than once. When the distance between two parallel mirrors is minimized, the divergence will be maximized due to a larger angle between the possible paths of reflected light.
In a very basic laser that isn't designed to produce single mode output, the laser beam is not a cylinder of evenly distributed light. The light takes many paths through the resonator as it is reflected by the mirrors.
In the case of lasers that use windows, openings or holes to pass the output, light will never make it through the window if the rays are truly parallel. Aside from the fact that a beam of light is never perfectly parallel, a complex pattern of modes ensures that light will always escape from the window. Beam quality is very poor.