The performance of a HTTP proxy is limited by the performance of the used disk subsystem. The proxy needs to write a new object to the disk subsystem in case of a Cache MISS (the object is new and was not stored in the cache yet). In case of a Cache HIT (the client requested an object that is already stored in the cache) it needs to read the object from the disks to serve it to the client. That means that each HTTP request results into a request to the disk subsystem in the worst case. Caching some parts of the requests in RAM helps to reduce the load to the disks a little bit but does not change the performance dramatically.
Each enterprise SATA disk with 7.2k rpm can only handle approximately 100 requests per second. Faster SAS disks with 15k rpm can handle approximately 200 requests per second. That means that you need 30 SATA or 15 SAS disks to reach a performance of 3000 requests per second.
CacheMARA uses high end SSDs to dramatically increase the performance. SSDs do not have any moving parts and have ultra low access times. High End SSDs can do fifty thousand and more requests per second. The use of SSDs allows CacheMARA to perform at many more requests per second than any other caching solution on the market.
You should not trust any vendor that tries to tell you that a HTTP proxy system that only uses SATA or SAS disks can do much more than 100-200 request per disk per second. This is just impossible because of the physical limitations of the disks. Disks have moving parts and it takes time to move the head to the requested position on the disk and read or write the object.