The core S3 API is beautiful

But that same simplicity makes it unsuitable for some things.

Swap-to-S3 should totally be viable, but the protocol overhead for small requests largely makes it useless for that.

Likewise, it's very hard to use hardware acceleration to construct read/write requests to S3, or DMA an S3 response direct into RAM.

Sure, S3 isn't designed for that stuff, but the protocol shouldn't be the limiting factor.

The beauty isn't in the UX of the API per se, it's in the reduction of historically complex file system APIs down to a simple set of primitives and behaviours that enable massive throughput and scale while still being broadly applicable to most modern application needs.