How much code (and how slow is it) reads in a directory and then stat's
every object?  How much could be gained by having a system call which
read in a directory and returned an array of names and stat structs?

Or even another thought might be to schedule (at a very low priority,
and one which allows requests to be thrown away before they get run)
reading in dinodes for all entries in a directory after the directory
contents have been read.

This maybe really stupid (because the names and dinodes don't co-exist
in the same space, on disk) but it can't hurt to ask others about what
they think of this :)  Heck, they're probably really dumb ideas but it
can't hurt to share them, can it ? O:-)

What prompted me to ask?  Reading someone saying write "Konquerer took 
forever to display a large directory on Linux" and then say "Explorer
windows popped up almost instantly".

Darren