I recently went through this myself as well. Considering an in memory cache will be process specific (not shared across multiple instances of a website or native business app or multiple servers) there is really no benefit to having multiple MemoryCache instances except for code organizational reasons (which can be achieved in other ways).
The Memory cache is intended to be used alone mostly because of its memory management capabilities. In addition to the performance counters (which do have some overhead) the MemoryCache is also able to expire items when it runs out of allocated memory.
If the current instance of the cache exceeds the limit on memory set
by the CacheMemoryLimit property, the cache implementation removes
cache entries. Each cache instance in the application can use the
amount of memory that is specified by the CacheMemoryLimit property.
from MemoryCache.CacheMemoryLimit Property
By using only one instance of the MemoryCache it can apply this memory management efficiently across the entire application instance. Expiring the least important items across the entire application. This ensures maximum memory use, without exceeding your hardware capabilities. By limiting the scope of any one MemoryCache (like to one instance of a class) it can no longer effectively manage memory for your application (as it can’t “see” everything). If all of these cache’s were “busy” you may have a harder time managing memory and it will never be nearly as efficient.
This is particularly sensitive in applications which don’t have the luxury of a dedicated server. Imagine you are running your app on a shared server where you’ve only been allocated 150mb RAM (common cheap $10/month hosting) you need to count on your cache to use that to the max without exceeding it. If you exceed this memory usage your app pool will be recycled and your app loses all in memory caches! (common cheap hosting practice) The same could apply to a non-web app hosted in house on some shared corporate server. Same deal, you’re told not to hog all the memory on that machine and to peacefully co-exist with some other line of business apps.
That memory-limit, app pool recycle, lose caches thing is a common “Achilles heel” to web apps. When the apps are their busiest, they reset the most often due to exceeding memory allocations, losing all cache entries and therefor doing the most work re-fetching stuff that should have been cached in the first place. Meaning the app actually loses performance at max load instead of gaining.
I know MemoryCache is the non-web specific version of System.Web.Caching.Cache implementation, but this illustrates the logic behind cache implementation. The same logic can apply in a non-web project if you don’t have exclusive use of the hardware. Remember if your cache forces the machine to start doing pagefile swaps then your cache is no longer any faster than caching on disk. You’ll always want a limit somewhere, even if that limit is 2gb or something.
In my case after reading up about this, I switched to using one ‘public static MemoryCache’ in my app and I simply segregated cached items by their cache keys. For example if you want to cache on a per instance you could have a cache key like something like “instance-{instanceId}-resourceName-{resourceId}”. Think of it as name spacing your cache entries.