homepage

I wish I shared your optimism about this, but I don't. Here's a short 
explanation why.
The problem of I/O and the associated interface between hardware, the 
operating system kernel, and user applications is one of the most 
fundamental and carefully studied problems in all of computer systems. 
The C library and its associated I/O functionality provide the user-
space implementation of this interface. However, if you peel the covers 
off of the C library, you're going to find a lot of really hairy stuff 
in there. Examples might include:
1. Low-level optimization related to the system hardware (processor 
architecture, caching, I/O bus, etc.). 
2. Hand-written finely tuned assembly code. 
3. Low-level platform-specific system calls such as ioctl().
4. System calls related to shared memory regions, kernel buffers, etc. 
(i.e., optimizations that try to eliminate buffer copies). 
5. Undocumented vendor-specific "proprietary" system calls (i.e., 
unknown "magic").
So, you'll have to forgive me for being skeptical, but I just don't 
think any programmer is going to sit down and bang out a new 
implementation of buffered I/O that is going to match the performance of 
what's provided by the C library.
Again, I would love to be proven wrong.