larger buffer.

* Hand code a reverse string join.
* Do not use defer.

Speed up: 7%

the request to be reused, thus reducing garbage generation.
    
Back to old speed (10% slower than libfuse).

Add test for memory pressure. This brings performance back to 50%
slower than libfuse, but simplifies structure for future performance
improvements.

the request to be reused, thus reducing garbage generation. 

Speed-up of ~35%.  Go-FUSE is now only 10% slower than libfuse.