If this is for a root server hosted in AWS, then it might make sense. That said, I would probably only do it for the root server, not the top level relays so it isn’t shared by multiple systems trying to maintain the cache.
If the networking between the root and the top level relays is fast enough, then the size of the cache probably only needs to be big enough on the relays so that it can hold about 72hours worth of content so the caching doesn’t roll over too fast.
For the root server, you potentially want the cache to be as big as possible and raw IOPS and low latency for the cache are not as important on the root server if all the downloading is actually being done through the relays themselves. I think ideally you want the root server to be able to hold 30 to 180 days worth of cache. I would generally recommend not less than 100GB if you are heavy user of patch and software distribution. The bigger the better typically.