hg branches performance regression

[Steven Brown]

This changeset makes "hg branches" slower on a branchy repo:

changeset:   16439:e5750c6716eb
user:        Matt Mackall <mpm at selenic.com>
date:        Fri Apr 13 21:35:48 2012 -0500
summary:     revlog: increase readahead size

Before:
$ time ./hg branches -R ../10k > /dev/null
real	0m4.225s
user	0m3.132s
sys	0m0.960s

After:
$ time ./hg branches -R ../10k > /dev/null
real	0m19.068s
user	0m3.768s
sys	0m15.265s

where ../10k is a local clone of https://bitbucket.org/mg/10k with an
existing .hg/cache/branchheads file.

Caching the closed state of the branch heads fixes this, but I don't
yet have a finished patch ready, and it doesn't seem appropriate for
the stable branch anyway.

Is it OK to back out this change for the 2.2 release and revisit it afterwards?