The attached diff has my change to aplay.c:capture_go and supporting globals, 
to implement a two-thread double-buffering approach to capture.  I kept having 
overruns, due apparently to file system lock contention in RedHat 9 (2.4.20-8).

The approach is to allocate a 10-second-long buffer (rounded up to a multiple 
of chunk-size), start a disk-write thread to follow behind the original 
thread, and if the file system lock causes the write to block for a while, 
capturing continues unabated.  10 seconds has been sufficient so far; the 
longest FS lock I could cause was 9.5 seconds.  No overruns (with a big enough 
buffer-size) since this patch has been in use!

I release the IP in this patch to the ALSA project for inclusion, should the 
pertinent community members find it worthwhile.

Cheers...

Eric Weaver
Palo Alto, CA