TCP_SWAPFAIL_MISS and asyncufs/2.3

We have just moved to squid-2.3.STABLE2 with "--enable-snmp
--enable-async-io" re-using a squid-2.3.STABLE4 spool, cache_dirs
configured as asyncufs, and have seen our object and byte hit rates drop to
the floor. With 2.2 we had about 25-35% byte hit rates, but now it is
around 0-1% and cache.log is showing a lot of

2000/04/05 02:32:45| WARNING: swapin MD5 mismatch
2000/04/05 02:32:45| storeClientReadHeader: URL mismatch
2000/04/05 02:32:45| {http://www.asianmail.org/dt/dir.gif} != {http://209.132.206.17/vvtour/images/tours1_r4_c2.gif}

along with a LOT of TCP_SWAPFAIL_MISS entries in access.log.

Presuming object incompatibilities between the two versions I newfs'd and
re-built new empty cache dirs (swap.state lives on these disks also, so
they are cleared), and started 2.3 back up. We are still seeing a
lot of TCP_SWAPFAIL_MISS (mixed between /200 and /304 mainly) in
access.log - roughly 3 times as many entries as TCP_HITs.

Q's:
1. TCP_SWAPFAIL_MISS means squid thought it had the object but cannot
retrieve it from the cache. Why would this be happening (especially on a
fresh cache)?

2. Are 2.2 and 2.3 cache spools supposed to be compatible? (and does
asyncufs make any difference to compability?)

3. is asyncio stable enough under 2.3? squid performance appears to be
much improved with it, only problem is the hit rates.

note: OS is Solaris 2.6, running on UltraSparc.

CM

-- 
Chris Miles  -  System Architect  -  connect.com.au
cmiles@connect.com.au                     Australia
http://www.psychofx.com/chris/       +613 9251 3610