Decided to play with differing block/fragment sizes and cpg on
	a new 17GB disk on my Ultra1 (1.5Y, dirpref, softdep, etc).

	test was to untar an uncompressed tarfile of the source tree,
	note the disk usage, then rm, then umount.

	Looks like on this system 16/2k and doubled cpg is about 3%
	faster than the standard 8/1k, and 32/4k is slower than the
	default.

	Not terribly scientific (should have varied cpg independently
	to block/fragment as well), but I needed to get the disk newfs()ed
	and running :)

	Interesting df numbers...

	    cpg                18      64     128
	    cylinder-groups   417     117      58
	    block(bytes)     8192   16384   32768
	    frag(bytes)      1024    2048    4096

	    untar(sec)        349     337     344
	    rm(sec)           346     335     401
	    umount(sec)        97      98     101

	    df-total(K)     16497   16759   16889
	    df-used(K)        312     347     418
	    df-free(K)      15360   15573   15627

	Note: cylinder-groups calculated by hand afterwards.

-- 
		David/absolute		abs@formula1.com