IO related interactivity regression in recent -current

hi

in recent weeks I am observing sudden drop of interactivity which
is related to IO. thats is - when one process does IO and another
tries to issue some IO as well the second one is terribly starved.

nice testcase is:

long running compilation in 1 window

pkg_version -vl "<" in another window

the pkg_version takes A LONG time to finish

example:

(on idle system)
witten ~# time pkg_version -vl "<"
....
21.336u 13.078s 0:32.88 104.6%	320+582k 0+0io 0pf+0w

(on gmake -j2 compilation of llvm)
witten ~# time pkg_version -vl "<"
....
23.275u 10.937s 3:20.62 17.0%	288+506k 186+0io 121pf+0w


ie. it takes 7times takes that long to finish.

can someone shed some light on this? confirm this?

thnx

roman