Yes, I just experienced this as well.  And if you include doing
`make depend' it gets even worse, I think.  This code cannot
stay as it is, I mean a 3-fold slowdown is pretty awful.

	-- Lennart

Kouichirou Hiratsuka wrote:
> Hello,
> 
> I updated my system (kernel and userland) today. It works well for a
> while, but performance of compile drops by some disk access.
> 
> # cd /usr/src/sys/arch/i386/compile/GENERIC
> # make clean > /dev/null; time make > /dev/null
>       381.73 real       339.84 user        34.39 sys
> # /usr/libexec/locate.updatedb
> # make clean > /dev/null; time make > /dev/null
>      1043.92 real       342.24 user        45.92 sys
> 
> My system has 1G bytes RAM and free memory was about 550M bytes at
> the time of second build of the kernel. But it was very slow by hard
> disk accesses. It seems to be related with new memory allocation
> scheme for the buffer cache.
> 
>   http://mail-index.netbsd.org/source-changes/2003/12/30/0011.html
> 
> It did not happen before this commit.
> 
> best regards.
>