Linux openSUSE 13.1 x64, KDE 4.14.1. Radeon HD 6870, FOSS video driver (Gallium / Mesa).
Short description of error
When using multiple simulation caches per cloth or particle system, only the cache that's selected in the Particle / Physics -> Cache window appears to be evaluated. The problem with this is that, if you have multiple caches where each represents a different frame range, and wish to render all of your scenes in one go, you won't have all hair / cloth physics evaluated automatically.
Multiple caches per cloth or particle system are most necessary when an object reappears in a different scene (Blender scene or a different camera angle) and must instantly appear in a different location over the course of one frame. If an object suddenly reappears somewhere else, the cloth / particle physics interpret that as the object having moved extremely fast, and go all over the place during the first frames after the movement. For this reason, you want to have one cache that ends before the object moves, and another cache that starts the next frame as the object has moved. Of course, you can avoid this by using a duplicate of your object in the second shot, but that's less optimal.
Exact steps for others to reproduce the error
1 - Create any object that contains a particle system or cloth simulation.
2 - Assuming the scene starts at frame 1 and ends at frame 300: Create two caches under the Physics / Particle window. One starts at frame 1 and ends at frame 150, the other starts at frame 151 and ends at frame 300. Set them both to Disk Cache then bake both.
Issue - Only the cache you select is visible... at least in the viewport, and I assume during rendering as well. The other cache is ignored.
Solution - If this is not the intended behavior, I assume it's a bug that needs to be fixed. If it's intended, I suggest adding an option to interpret all caches on an object simultaneously, so that both are visible in the viewport and respected at render time. If two caches overlap (one ends after another starts), the bottom one should have priority.