Hi Charles,
ImageIntegration's cache is managed automatically; it requires a very small amount of RAM to store image statistical data and cannot be the cause of an out-of-memory error (unless you run under really severe memory constraints, but in such case PI probably wouldn't even run, anyway).
The probable cause is the amount of RAM required by all pixel buffers. There is an independent buffer for each integrated image. The default buffer size is 16 MB so with 100 images you're needing in excess of 1.6 GB (which, considering suboptimal memory allocation, may easily require 2 GB of effective allocated space). If you're running a 32-bit version of Windows, then here's the problem. Note that in such case your 4 GB are not available to applications due to the default Windows 32-bit memory management policy.
Try lowering the Buffer size parameter; for example, try setting it to 8 or 4 MB. If you enter zero for this parameter, then ImageIntegration will use just the space required to store one row of pixels, which is the absolute minimum. This may have a severe impact on performance though, as it will increase the frequency of disk accesses and will degrade the performance of multithreaded pixel rejection routines, but will probably let you work.
