I have not paid much attention to the ACR cache, so this question made me curious. Curious enough to do some research and some testing.
Google research led to a lot of bad info, mostly due to confusion between the Bridge cache, the Photoshop cache, the Lightroom cache. A lot of "experts" seem to think the cache contains previews. I don't think that's true.
I finally found this site: http://www.pixiq.com/article/camera-raw-cache
That link contains the following quote, "The early part of the process includes decoding, decompressing, linearizing, hot pixel filtering, and demosaicing an image. The results of this are the same each time, so for this reason, it’s perfectly legitimate to use a cached version of this as the basis for a preview."
That makes good sense, but the key words there are "as a BASIS for a preview." The actual preview you see in ACR has to have the ACR defaults applied. At a minimum, the true preview has to have a camera profile applied.
My first simple test was to bang away in ACR with adjustments on a single raw file. No matter what you do in ACR, the ACR cache file never changes after its initial creation. Seems to support the above claim that the cache file contains only the "basis." Also, note that if you open a tif or jpeg in ACR, no ACR cache file is created (so where does the ACR preview come from?)
My next test was to purge the cache, then open 8 raw files in ACR and look at the sizes of the cache files. The files were from a Canon 5D2 (5616x3744 pixel images). They cache files varied from 246kb to 107kb. Correlating the cache files to the images, it appears the differences are due to the level of detail in the images. Lower detail equals lower cache file size. So, it appears there is some kind of compression going on. Just like with jpegs, lower detail equals greater compression and smaller file size.
Then I purged the cache and loaded 8 images into ACR again, timing the result with a stopwatch. It took 5 seconds for all the thumbs to appear in ACR and for all the yellow triangle explanation marks to go off. Then I simply did it again with the previous cache in place. This time it took 4 seconds.
Even given a generous degree of error due to such a small sample and manual stopwatch timing, that's not much savings. And given that most of the time you are opening one image at a time, the time saved by the cache is a small fraction of a second. So, my conclusion is that the ACR cache is of very little value. Why waste several gigabytes of disk space on it?
But then a great mystery occurred. In previous tests I opened 8 images at a time by selecting them in Bridge and opening in camera raw. I purged the cache again and opened some of these same images one at a time. When opened one at a time, the cache file created is larger than when opened in a batch. A lot larger. An average of 2.6 times larger, with little variation in that ratio. And there is more mystery. When opening a batch of images, the cache for the first image is the same size as when it's opened alone. Only the 2nd through the last images get smaller cache files than when opened alone.
I can't figure out that mystery. Would appreciate someone else doing the test to confirm.
All this is on Win7 64bit with CS6 and ACR 7.1.